Expert Systems With Applications 194 (2022) 116548

Contents lists available at ScienceDirect

Expert Systems With Applications

journal homepage: www.elsevier.com/locate/eswa

Decentralizing construction AI applications using blockchain technology

Kareem Adel a, *, 1, Ahmed Elhakeem a, 2, Mohamed Marzouk b, 3
a
Construction and Building Engineering Department, College of Engineering and Technology, Arab Academy for Science Technology & Maritime Transport (AASTMT),
Cairo, Egypt
b
Structural Engineering Department, Faculty of Engineering, Cairo University, Giza, Egypt

A R T I C L E I N F O A B S T R A C T

Keywords: Decentralizing the Artificial Intelligence (AI) applications is deemed the next wave of Blockchain Technology
Blockchain (BT) in the construction industry. Most of previous research studies employed the AI and the BT separately with
Trusted Decentralized AI limited efforts providing a conceptual view for possible convergence in real-life applications. Nevertheless, such
Chaincodes
efforts do not address construction applications. This research introduces a novel tailorable decentralized AI
Decentralized Inference
Decentralized Learning
system that utilizes BT as a computing-oriented technology. The proposed system is developed as an inference
Construction Cost Estimate engine while having a number of interesting features. First, the system validates and audits the decision-making
process while sharing and recording the input data and the computed outcomes in a synchronized trusted
manner. Second, the system allows the formation of distributed AI repository that can absorb and manage
concurrent use-cases while targeting different scopes and covering diverse AI branches. Third, it provides a
workable solution for the AI applications’ distribution problem, which hinders their wide employment. Fourth,
the introduced system guarantees sustainable versioning and evolution over time for AI applications based on
their performance or the newly acquired data. A case study of estimating the construction cost for road projects is
provided to illustrate the system’s workability and demonstrate its performance.

1. Introduction Pandl et al., 2020; Research-and-Markets-Store, 2020; Vyas et al., 2019).

Most of construction research efforts focus on utilizing AI models
Blockchain Technology (BT) and Artificial Intelligence (AI) are the separately to add value or transform the industry processes depending
most hyped developments in information technology. These technolo­ on a centralized approach starting from model feasibility, data acqui­
gies deemed as a major engine to revolutionize the construction industry sition, and model development to model deployment. Such approach
since it bears many misalignments, fragmentation, and low productivity suffers from three major issues (Ekramifard et al., 2020; ElMousalami
and effectiveness. AI can provide computers and machines with cogni­ et al., 2018; Gupta, 2020; Pandl et al., 2020; Salah et al., 2019). Firstly,
tive capabilities to learn, infer and decide based on a collection of data the provenance and authenticity of data sources are not warranted. As a
and sometimes, these capabilities can exceed the human ones. Accord­ result, the data becomes questionable, and the outcomes of developed
ingly, a report by McKinsey Global Institute in 2018 stated that AI ap­ models could be biased or mistaken. Secondly, the centralized owner­
plications would supply an additional global economic value around ship of data exposes it to undesirable risks like single-point attacks,
USD 13 trillion by 2030. On the other side, BT can provide consensus tampering, or leakage of sensitive records. Thirdly, the researchers or
over a digital distributed ledger in an untrusted environment that may developers exert repetitive efforts for reproducing same purpose AI
consist of unreachable nodes or stakeholders while securing access models due to the absence of a secure sharing platform/system for the
control for electronic records and IT systems. Based on that, a study by pre-existing ones.
Research and Markets Store expects that the global BT market size On the other side, recent research efforts have investigated the usage
would grow from USD 3.0 billion in 2020 to USD 39.7 billion by 2025 of BT in the construction field. These efforts can be clustered into four
with 67.3% as a compound annual growth rate (Dinh & Thai, 2018; major domains; payment management (Elghaish et al., 2020; Ye &

* Corresponding author.
E-mail addresses: [email protected] (K. Adel), [email protected] (A. Elhakeem), [email protected] (M. Marzouk).
1
ORCID: 0000-0002-8193-0204.
2
ORCID: 0000-0001-6585-8823.
3
ORCID: 0000-0002-8594-8452.

https://doi.org/10.1016/j.eswa.2022.116548
Received 2 January 2021; Received in revised form 1 December 2021; Accepted 12 January 2022
Available online 21 January 2022
0957-4174/© 2022 Elsevier Ltd. All rights reserved.
K. Adel et al. Expert Systems With Applications 194 (2022) 116548

König, 2020), chaining of quality and building information (Sheng, Machine Learning (ML) is the most exploited AI branch in the con­
Ding, et al., 2020; Xue & Lu, 2020), supply chain management (Helo & struction engineering and management field for diverse purposes. These
Shamsuzzoha, 2020; Wang et al., 2020) and decentralized autonomous purposes include cost/price estimating, productivity measurements,
organizations (Hunhevicz & Hall, 2020; Sreckovic & Windsperger, assets’ inspection and maintenance, construction safety and health
2019). The efforts have employed the BT as a transaction-oriented management, and risk analysis for construction projects (See Table 1).
technology for only exchanging and maintaining raw textual and ML focuses on developing applications and programs that can learn via
numeric data without conducting real-time analytics or further analysis experience. These applications enclose a model that captures the linkage
based on the whole chained data. between input data and output data. Such a model has to be trained
The convergence of AI and BT in a single ecosystem is considered a using supervised or unsupervised learning algorithms depending on its
perfect mechanism to overcome the issues mentioned above, leading to scope. The ML model’s inner working is described as a black box since its
what can be called a Trusted Decentralized Artificial Intelligence (TDAI) functionalities and outcomes are intricate to be explained which reduces
(Corea, 2019; Gulati et al., 2020; Harris & Waggoner, 2019; Inbaraj & the dependability upon it (Nassar et al., 2020).
Chaitanya, 2020; Tanwar et al., 2019). TDAI provides a stable distrib­
uted medium to secure the AI models’ sharing without the need for in­
2.2. Blockchain
termediaries or trusted third parties. TDAI also enables securing,
auditing, and validating the learning data to avoid the development of
BT represents a significant branch of distributed ledger technologies
mistaken or biased AI models. Furthermore, TDAI via the utilization of
that encloses a digital ledger that scattered and being updated around
smart contracts or chaincodes in BT allows the outdated AI models to be
multiple nodes in a synchronized distributed manner. These nodes form
modified, versioned, and evolved over time based on validating and
a peer-to-peer network that enables direct interaction and information
calibrating their outcomes by the relevant stakeholders. Therefore, if the
exchange without the need for any trusted intermediaries while
outcomes are reliable, the models are still shared and embedded in the
ensuring security and reliability for all nodes (Elghaish et al., 2020). The
chaincodes. Otherwise, access to outdated versions is restricted, and
inner working of BT depends on a list of blocks that are chained together
new valid versions are re-developed using off-chain or on-chain learning
using unique cryptographic hashes. These hashes allow the blockchain
algorithms based on the chained new data.
transactions to be traceable and impervious against change or tampering
The research addresses this potential by developing a decentralized
(Bamakan et al., 2020; Papakostas et al., 2019; Xu et al., 2019).
AI system that can be tailored to suit various concurrent applications
within the construction companies, research centers, or labs. The pro­
2.2.1. Blockchain as a Transaction-Oriented technology
posed system answers the research question of how a decentralized BT
BBN is mainly used as a distributed data structure that able to
system can enhance the trust in AI models’ decisions, and analytics. The
circulate and maintain diverse types of data with a high degree of
system’s scope is to build a private blockchain-based network (BBN) that
encryption and transparency while providing no centralized manage­
encloses a digital AI models’ library. This library is coded within a
ment since all nodes have the same rights and obligations (Ar et al.,
chaincode to facilitate data transaction and storage while providing
2020; Kim et al., 2020; Penzes et al., 2018; Tang et al., 2019). Thus, it is
decisions or predictions based on such data via a Command Line based
utilized to minify business frictions, reduce operational costs, settle
(CLI-based) interface. The CLI-based interface is utilized mainly for
transactions’ vulnerability, and provide a trustworthy information ex­
prompt implementation and testing at a more controlled and granular
change system. BBN is classified into permissioned-less network and
level. The proposed study contributes to the body of knowledge since it
permissioned network. The permissioned-less network can be joined by
presents a novel decentralized intelligent system for construction com­
any participant with no restrictions while being secure due to its
panies and research centers to exchange and analyze information using
powerful cryptographic algorithms. On the other side, the permissioned
blockchain and AI in tandem. Also, it provides practical guidelines for
network is mainly used to build a platform for a specific organization
the development and the implementation of such a system in the con­
that can be accessed by pre-identified or selected participants (Penzes
struction industry that is characterized by different parties with
et al., 2018). According to Chang et al. (2019); Penzes et al. (2018);
numerous communication channels
Perera et al. (2020), the major features of these types are summarized as
The remainder of this paper is structured as follows: Section (2)
follows:
presents theoretical background related to AI and BT. Section (3) clar­
ifies the research methodology for the proposed system. Section (4) fo­
• Security: BT uses impenetrable encryption mechanisms to assure the
cuses on the system implementation. Section (5) is related to the testing
validity of stored data and prevent manipulation.
stage to verify the system’s capabilities. Sections (6) and (7) provide
• Transparency: BT orders and sends the transactions to all members
thorough discussion and conclusions.
in an apparent way.
• Decentralization: BT allows (Write-Once-Read-Many) characteris­
2. Theoretical background
tics with no localized data storage or directory to reduce delays and
avoid a single point of failure.
2.1. Artificial Intelligence
• Auditability: BT time-stamps the on-chain data to be easily traced
by different nodes.
AI is an advanced computational discipline that enables machines
and computers to perform complex tasks based on a knowledge database
Table 1
to make decisions, predictions, or classification (Krittanawong et al.,
Recent AI Studies in the CEM Field.
2020). It can be sorted into Artificial General Intelligence and Narrow
Artificial Intelligence. Artificial General Intelligence refers to self- Category Reference

control computers that able to solve tasks in a wide range of different Construction Safety and Baker et al. (2020); Cheng et al. (2020); Choi et al.
domains. In contrast, Narrow Artificial Intelligence refers to AI-based Health Management (2020); Sakhakarmi et al. (2019)
Built Assets’ Inspection and Guo et al. (2020); Maeda et al. (2020); Pan and Yang
applications that are built to solve a particular task in a specific nar­
Maintenance (2020); Yao et al. (2020); Zhu et al. (2020)
row domain. Most of existing AI applications and systems rely on Nar­ Risk Analysis El-Kholy (2019); Fan (2020); Gondia et al. (2020);
row Artificial Intelligence (Pandl et al., 2020). The divisions of Narrow Sanni-Anibire et al. (2020)
Artificial Intelligence include but are not limited to Expert Systems, Cost/Price Estimating Cao et al. (2018); El-Kholy et al. (2020); Gondia
Machine Learning, Deep Learning, Natural Language Processing, and et al. (2020); Matel et al. (2019); Rafiei and Adeli
(2018)
Robotics (Kim & Huh, 2020; Panda & Jena, 2020; Pandl et al., 2020).

2
K. Adel et al. Expert Systems With Applications 194 (2022) 116548

Despite these features, some issues like network type and size, block 2.3. BT and AI convergence
time and size, storage bloating, transaction throughput, and latency still
need technical enhancements. BT can amplify the conventional AI systems’ capabilities by enabling
a decentralized digital ledger that performs consensus and provides
2.2.2. Blockchain as a Computing-Oriented technology agreed-upon decisions. These decisions are stored on blocks to be later
BBN can be utilized as a general computing platform through the on traced back and audit. It also allows AI systems to act as autono­
assembly with Smart Contracts (SCs) or Chaincodes (CCs) that provide mously driven-based systems that can collect, store and analyze data.
the BBN with the needed computational power and programming logic. According to Dinh and Thai (2018); Ekramifard et al. (2020); Gulati
Although smart contract or chaincode seems like a new notion, it was et al. (2020); Harris and Waggoner (2019); Inbaraj and Chaitanya
proposed in 1994 by Szabo (1994). SC or CC is defined as a self- (2020); Mamoshina et al. (2018); Nassar et al. (2020); Pandl et al.
executing program that can be triggered when specific terms and con­ (2020); Salah et al. (2019); Sarpatwar et al. (2019); Sgantzos and Grigg
ditions are fulfilled while minimizing exceptions and intermediaries (2019); Tanwar et al. (2019); Vyas et al. (2019), the benefits of utilizing
(Nassar et al., 2020; Tanwar et al., 2019). The evolution of BBN affords a the BT for AI can be summarized as follows:
low-budget path for building and deploying CCs. Since CC contains
trigger conditions and relevant responses, it can be embedded in the • BT can provide a digital diskless environment to process AI models/
BBN to serve as a state machine that constantly observes data fluctua­ architectures
tions in the chain while automatically execute predefined actions. CC • BT can ensure data provenance, authenticity for the training algo­
can also handle complex programmable scenarios by using sets of rich rithms to learn, infer and make credible outcomes or decisions.
functions that are available and supported by Turing complete lan­ • BT can enhance the transparency and the trust to understand and
guages, such as Go, JavaScript, and TypeScript for Hyperledger Fabric audit the AI models/architectures outcomes without the need for
and Solidity for Ethereum (Perera et al., 2020; Sgantzos & Grigg, 2019). third-party auditors.
With such unique capabilities, many researchers are recently • In AI Multi-user processes that involve various parties such as busi­
exploring BBN-CC integration in domains as diverse as healthcare ness firms and governmental authorities, BT can automate and
management, supply chain management, banking and financial services, accelerate the decision-making and the data validation between
e-voting, Internet of Things (IoT), distributed access control, and energy these parties.
supply. However, some challenges like lack of standards and protocols, • Since AI models/architectures subtend some reasoning issues due to
legal issues, looping issues, and error intolerance are deemed the their black-box behavior, BT can help to better explain those models/
bottleneck that hinders the technological development and employment architectures by providing a clear route to trace back the develop­
of CCs (Hasan & Salah, 2018). ment pattern during their life cycle and AI decision-making process.
• BT and CCs can make AI models/architectures perform more smartly
2.2.3. Blockchain applications in construction by making near-instantaneous engagements “real-time analytics” or
Recent research efforts have presented the potential uses of BBNs in business value engagements “analysis based on historical data”.
the construction field, which is deemed a vital step to suit more detailed • BT can lower the market barriers for newly developed AI models/
applications. These efforts focus on employing the BBNs as a transaction- architectures entry since they are developed based on tamper-proof
oriented technology for exchanging raw textual and numeric data. The data while using a decentralized direction.
efforts can be classified into four major categories as listed in Table 2. • BT can restrict or minimize risky scenarios since on-chain AI models/
The first category is payment management in which BBN has utilized to architectures only process specific actions and nothing more.
control, secure, and direct the financial invoices for ongoing construc­
tion projects in an automated approach. The second category is the The emerging research trend is directed toward using BT as a
chaining of quality and building information, which can be a perfect computing-oriented technology for decentralizing AI operations, espe­
solution for the information fragmentation issues via providing a single cially model data storage and management, model development, and
source of truth platform for the construction industry while ensuring model deployment.
transparency and auditability. The third category considers the use of
BBNs to upgrade supply chain management by ensuring the origin and 2.3.1. Data storage and management
the timing for procurements and submittals in a verified effective The current centralized storage mechanisms are considered highly
approach for all parties. Finally, BBNs can fully manage and direct susceptible to privacy and security issues and inefficient with regard to
construction enterprises without any human mediation via CCs that hold massive data transfers or retrievals. As a result, Blockchain-based stor­
the governing rules and conditions, which leads to decentralized age mechanisms can secure the sensitive data and minimize the latency
autonomous organizations. of AI processes via enabling parallel data access from different
geographical network nodes (Corea, 2019; Dinh & Thai, 2018; Ekra­
mifard et al., 2020; Gupta, 2020; Inbaraj & Chaitanya, 2020; Pandl et al.,
2020; Salah et al., 2019; Tanwar et al., 2019).

Table 2 2.3.2. Model development

Recent BT Studies in the CEM Field. Model development refers to the training, testing, and validation of
Category Reference AI models. Traditionally, it is performed using a centralized approach
before real-world deployment. This approach is considered inefficient
Payment Management Ahmadisheykhsarmast and Sonmez (2020); Chong
and Diamantopoulos (2020); Das, Luo, et al. (2020); since it blocks the continuous evolution of models themselves and may
Elghaish et al. (2020); Ye and König (2020) provide outcomes based on old data. Decentralizing the model devel­
Chaining Quality and Cerić (2019); Das, Tao, et al. (2020); Hijazi et al. opment is a feasible solution that would lead to highly customized
Building Information (2019); Li et al. (2020); Nawari and Ravindran learning models for each client or purpose and leverage the continuously
(2019); Sheng, Ding, et al. (2020); Sheng, Luo, et al.
(2020); Xue and Lu (2020); Yang et al. (2020); Zhang
evolving data streams. Moreover, it allows periodical or incremental
et al. (2020); Zhong et al. (2020) models’ building or tuning since it avoids starting the training process
Supply Chain Management Helo and Shamsuzzoha (2020); Wang et al. (2020) from scratch. This can be done by shielding and scaling the computa­
Decentralized Autonomous Hunhevicz and Hall (2020); Li et al. (2019); Shi et al. tional capabilities of smart contracts to effectively train, test, and vali­
Organizations (2019); Sreckovic and Windsperger (2019)
date the AI models on-chain (Dinh & Thai, 2018; Ekramifard et al., 2020;

3
K. Adel et al. Expert Systems With Applications 194 (2022) 116548

Gulati et al., 2020; Harris & Waggoner, 2019; Nassar et al., 2020; Pandl predefined privileges regarding write/read processes while holding its
et al., 2020; Salah et al., 2019; Tanwar et al., 2019; Wang et al., 2018). own ledger. The validation policy specifies the endorsement method for
a transaction that is proposed from one peer node to the other actors.
2.3.3. Model deployment The recording policy specifies the recording procedure of each validated
The model deployment or inference refers to the iterative execution transaction, including by which actor and through which channel. Such
of a trained AI model, which requires less computational resources than policy should be designed to succeed the actual relationships between
the model development (Tanwar et al., 2019). In the context of BT, the actors. The frontend access layer is proposed in the form of an
decentralized model deployment refers to embed or share the models as interactive CLI-based interface. This interface allows data feeding and
on-chain digital AI assets within the smart contracts. This approach can accessibility for authorized actors and outcomes’ traceability of AI op­
ensure the models’ provenance and authenticity while permanently erations. It can be directly linked with the blockchain layer by using
recording and tracking their changes and versions. Moreover, on-chain REST HTTP, which is an approach for communication with cloud data
models can be accessible and browsable, which forms a new channel centers.
for distribution and discoverability problems of new ones with the The system processes the data transactions in five main steps as
possibility to utilize them via Pay per Use option or Pay on Demand shown in Fig. 2. Firstly, an authorized actor uses the frontend interface
option (Bagchi et al., 2019; Pandl et al., 2020; Salah et al., 2019; to submit a transaction proposal for inferring a ML model. Secondly, the
Sgantzos & Grigg, 2019; Tanwar et al., 2019). interface communicates with BBN to activate the corresponding chain­
code to process the proposed inputs via the embedded ML model and
3. Research methodology provide an outcome. Thirdly, as predefined in the validation policy, the
inputs and outcomes are sent for validation to certain peer nodes.
This study introduces a novel decentralized AI system that performs Fourthly, once the validation is performed, the transaction data “inputs
as an inference engine for construction companies, research centers, and and outcomes” is time-stamped and stored within a block then broad­
labs. The system is tailorable to absorb concurrent ML models for diverse casted to the peer nodes to be committed and listed in their digital
use-cases. The architecture for such system is operated in a framework ledgers according to the recording policy. Fifthly, the same data is
that consists of three major layers: (1) Intelligent Chaincode Layer, (2) returned to the proposing actor via the frontend interface. At each
Permissioned Blockchain Network Layer, and (3) Frontend Access Layer transaction proposal, the same sequence is repeated.
as shown in Fig. 1. The intelligent chaincode layer is the key layer in
which computational operations and input-data processing are con­ 4. System implementation
ducted. The layer serves as a digital library that encloses off-chain
trained ML models for obtaining decentralized agreed-upon outcomes The proposed system was initially implemented for the case study of
while being responsible for storing the input data and such outcomes estimating the construction cost of road projects. Regarding the case
within the blockchain network and reporting them back via the frontend study, the key actors were cost engineer, decision/policy maker, and ML
access layer. The permissioned blockchain network layer includes the developer. These actors can be scaled at any point of time. The ML
key actors’ definitions, the transaction validation, and recording pol­ developer holds the overall governance to build ML models, configure
icies. Each actor involved in the network represents a peer node with BBN, technically train other actors and register them within the system.
The data exchange’s relationships between these actors were mapped as
shown in Table 3.

4.1. ML model

The ML model was developed for estimating the construction cost of

road projects in the early stages to assist the construction firms’
decision-makers in bidding and budgeting decisions. Seven parameters
were identified based on an in-depth review of the available database
from the General Authority for Roads, Bridges, and Land Transport in
Egypt. These parameters are listed as follows:

• Type of work (I1): The scope to be performed as a new road or a re-

pavement of an existing road.
• Grade of Road (I2): The functional grade of the road.
• Construction Duration (I3): The project duration in months for
completing the construction phase.
• Construction Year (I4): The difference between the construction year
and the base year which is 2002.
• Sector (I5): The district where the road is constructed.
• Road Length (I6): The total length of a road in Kilometers.
• Road Width (I7): The total width of a road in Meters.

The ML model was built as a multi-buffer perceptron artificial neural

network using NeuroSolutions Software Version 5.92. The model was
composed of processing units distributed among multiple buffers,
weights, and activation functions. The processing units represent the
computational nodes in which the input data is analyzed. The weights
represent the evolved learning embedded within the network. The
activation functions refer to the data interpretation functions used inside
each computational node (Cirilovic et al., 2014; ElMousalami et al.,
Fig. 1. Proposed Architecture for Decentralized AI System. 2018). The model’s topology was designated in three successive buffers

4
K. Adel et al. Expert Systems With Applications 194 (2022) 116548

Fig. 2. Dataflow for the System.

Table 3
Data Exchange’s Relationships.
Proposed by Transaction Transaction Validated by Ordered to
Type

Cost - Model Write/ - Decision Decision

Engineer Inference Submit Maker Peer Maker Ledger
Peer - ML Cost Engineer
Developer Ledger
Peer ML Developer
- Read Specific Read/ - None Ledger
Record Evaluate
- Query all
Records
Decision - Model Write/ - Cost
Maker Inference Submit Engineer
Peer Peer
- ML
Developer
Peer
- Read Specific Read/ - None
Record Evaluate
- Query all
Records
ML - Model Write/ - Decision
Developer Inference Submit Maker Peer
Peer - Cost
Engineer
Peer
- Read Specific Read/ - None
Record Evaluate Fig. 3. ML Model Structure.
- Query all
Records
Concerning the processing units’ number in the hidden buffer,
NeuroSolutions allows to be varied in a range then uses this range to
as depicted in Fig. 3 (Hegazy et al. (1994) and Gunduz et al. (2011)). build, train and cross-validate multiple models and automatically keeps
These buffers are: only the one with best results. Such range is determined heuristically
based on a trial and error process since there is no generic approach to
• An input buffer encloses seven processing units reflecting the input lean on as stated by Elbeltagi et al. (2014). In this study, the range was
parameters. set between 2 and 50 and the number of units with preferable results was
• A middle buffer consists of a number of processing units representing 38. The model relied on Levenberg Marquardt algorithm as a back-
the hidden buffer. propagation rule for supervised learning while utilized the hyperbolic-
• An output buffer consists of one processing unit representing the tangent function as a nonlinear activation function as recommended
average construction cost per (1 Km length, 1 m width & 1-month by Kim et al. (2004) and Hegazy and Ayed (1998).
Duration). The available database was divided into three sets [Training (60

5
K. Adel et al. Expert Systems With Applications 194 (2022) 116548

projects), Cross-Validation (5 projects), and Validation (10 projects)].

The size of each set does not rely on standard or acceptable generalized
rules as stated by Günaydın and Doğan (2004). However, each set
should cover all spectrums and variations of the input parameters. Both
training and cross-validation sets were involved in the ML model pro­
cessing with different functionalities. The training set was utilized to
learn and capture the cost relationships between input parameters and
output. In contrast, the cross-validation set was utilized for monitoring
the model’s performance to ensure the best level of generalization and
avoid the overtraining issue without affecting the network weights’
updating. The validation set was not used during the model processing
and was reserved aside to measure the validity and readiness of the
processed model to handle new or unseen cases.
It is worth mentioning that the hold-out method was used as the
cross-validation scheme as recommended by Adeli and Wu (1998) and
employed in previous relevant studies (Cho et al., 2013; Kim et al., 2013;
Kim et al., 2005). The model’s mathematical formulation and full results
are included as Supplementary Data. Table 4 summarizes the results for
each set in terms of Root Mean Square Error (RMSE), Mean Absolute
Fig. 4. Sample of Chaincode.
Error (MAE), and Mean Absolute Error Percentage (MAEP). As per the
table, the model estimates the cost with RMSE of 395.32, 148.53, and
(continued )
1413.13 EGP for training, cross-validation, and validation sets, respec­
tively. For MAE, the model provides predictions that deviate from actual Algorithm 1. Intelligent Chaincode
values by a mean of 240.44, 123.15, and 667.38 EGP for training, cross- Revert Inputs and Show an error
validation, and validation sets, respectively. Unlike the MAE, MAEP end
presents the predictions’ deviation from actual values as absolute per­ else if I1, I2, I3, I4, I5, I6, I7, ER Exists then
Execute ML Model to Calculate Construction Cost
centages equal 6.64%, 5.79%, and 9.73% for training, cross-validation,
Put Inputs and Construction Cost State on Chain
and validation sets, respectively. It is worth noting that the error per­ end
centages within the limits of parametric cost estimates set by AACEI else
(2020). Revert Inputs and Show an error
end
end
4.2. Intelligent chaincode layer Function (ReadAI)
if RId ∕
∈ ID then
Revert Inputs and Show an error
The chaincode was coded and packaged using IBM® VSCode end
extension, including three functions (InferAI, ReadAI, and QueryAI) as else
detailed in Algorithm 1 and shown in Fig. 4. The InferAI function holds Provide the Transaction with the given RId
the ML model and is used to calculate and provide an outcome for a end
end
specific project/record using a set of predefined parameters. This set can
Function (QueryAI)
be altered or scaled to accommodate the use-case under consideration. Provide All Transactions
Since the current case study is related to estimating the construction cost end
for roads projects, nine parameters were defined and listed as follows:

• [RId]: represents record number.

• [I1, I2, I3, I4, I5, I6, and I7]: represent cost estimating parameters. 4.3. Blockchain network layer
• [ER]: represents key actor judgment.
The blockchain network layer was established using IBM Blockchain-
The ReadAI function calls back a single record based on the [RId] as-a-Service solution (BaaS) that relies on Hyperledger Fabric as a
parameter, while the QueryAI function retrieves all on-chain records. hosting platform to leverage the benefits of permissioned BBNs, espe­
The chaincode sequence was configured as a closed-form script to avoid cially high performance, scalability, and API Support. IBM BaaS allows
looping issues or any possible logical errors. The next step is to develop different users to build, develop, interact and operate a permissioned
the blockchain network to install the chaincode on it. blockchain network within the context of their environment. Accord­
Algorithm 1. Intelligent Chaincode
ingly, the BBN was built via eight main steps as follows:
Step 1: a Kubernetes Service representing the computing cluster was
InputRId, I1, I2, I3, I4, I5, I6, I7, ER
created to manage the IBM BaaS and allow intelligent scheduling, re-
ID is the set of all Record’s IDs saved on the chain.
Function(InferAI) allocation, and balancing of computational resources.
if RId ∈ ID then Step 2: Certificate Authorities were created to provide the organi­
(continued on next column) zations with the required certificates for access and interaction.

• The organizations refer to the participants within the BBN and their
Table 4 numbers can be evolved over time without the need to rebuild the
Summary of Results.
whole network.
Set RMSE (EGP) MAE (EGP) MAEP (%)

Training 395.32 240.44 6.64% Step 3: Membership Service Provider (MSP) for each organization
Cross-Validation 148.53 123.15 5.79% was created to hold its definition and connection profiles.
Validation 1413.13 667.38 9.73% Step 4: The Peer nodes were created to manage the network ledgers.
*EGP: Egyptian Pound
Step 5: The Ordering Service was deployed to enroll peer nodes

6
K. Adel et al. Expert Systems With Applications 194 (2022) 116548

within its consortium. successive invocations via the CLI-based interface as shown in Figs. 5
and 6. The result shows acceptable computational speed for both writing
• The consortium is referenced as a list of the organizations’ peer and reading operations that nearly similar to conventional centralized
nodes that can build and run the validation and recording channels. systems as guided by Xu et al. (2019). The addition of new participants
reflects the system’s scalability feature that permits scaling the key ac­
Step 6: Channels were created to allow organizations to propose, tors without rebuilding the whole system. For the proposed system, This
validate and record transaction data within specific stream flows with process succeeds a simple three-step path that clarified as follows:
exposing such data to intended organizations only. Creating a Certificate Authority, a MSP and a peer node for the partic­
ipant and his/her organization, Updating the defined channel to enroll
• The channel has its ledger that is operated by the enrolled peer the participant in its consortium, Re-instantiating the intelligent
nodes. chaincode to take into consideration such participant. The incomplete
• The network can consist of several channels that could be built and data submission reflects the system’s ability to detect logical or pro­
enrolled by all or particular organizations and be used to run gramming errors before deploying the system. Therefore, when a
instantiated intelligent chaincodes. participant attempts to submit incomplete data, the system processes
such data via step 01 and step 02 as shown in Fig. 2. At this point, when
Step 7: Intelligent chaincodes that represent ML models were the corresponding chaincode function tries to handle the proposed input
installed and instantiated within the created channels. via the embedded ML model, it fails to derive an outcome. As a result,
Step 8: The connection profiles and the user identities for each MSP the system provides back a log-error message via the CLI-based interface
were created and downloaded to provide the CLI-based interface with that the input is incomplete, the computation is not conducted, and the
the required credentials to transact data effectively with the BBN. transaction is not recorded on the chain.

4.4. Frontend access layer 6. Discussion

The frontend access layer was configured based on the CLI-based The proposed research is a novel step toward providing a decen­
interface of IBM Cloud Function through using a serverless action. The tralized AI system for the different parties in the construction industry
action was coded using JavaScript language and operates as a stateless using the BT. The study holds four major theoretical contributions
Node.js code snippet. It was designed to interact with the BBN via the compared to relevant construction studies in expert and intelligent
MSP connection profiles while acting as a back-engine for the CLI-based systems. First, transforming the conventional AI systems to act as a
interface. The action invocation depends on the chaincode parameters persistent decentralized computing platform that automates and vali­
besides two additional parameters representing user identity [Id] and dates the decision-making process while sharing and recording the input
chaincode function [Fcn] as detailed in Algorithm 2. At this stage, the parameters and the computed outcomes in a synchronized trusted
system was fully developed and ready to be deployed. manner. Second, providing and forming a scalable distributed AI re­
Algorithm 2. Serverless Action pository. This repository is able to encompass concurrent models that
InputId, RId, I1, I2, I3, I4, I5, I6, I7, ER target different scopes (e.g., prediction, prioritization, classification and
KAID is the set of Key Actors’ IDs that saved in this code. optimization problems) while covering diverse AI branches like expert
Credentials are Chaincode-Name, Username, Key, Cert, and Connection Profile that systems, decision trees, search techniques and evolutionary algorithms
saved in this code.
unlike the previous studies in Table 1 that provided only non-tailorable
if Id ∕∈ KAID then
Revert Inputs and Show an error single use-case systems. Third, providing a workable solution for the
end distribution problem of AI models that reduces the dispensable repro­
else if Id ∈ KAID AND Fcn = InferAI then duction of similar purpose models. This feature tackles the shortcoming
Inherits Credentials in previous studies’ systems since their employment is confined to a
Execute Function (InferAI) in Algorithm 1
closed perimeter of developers or end-users. Fourth, guaranteeing the AI
end
else if Id ∈ KAID AND Fcn = ReadAI then models’ versioning and evolution over time which is considered a
Inherits Credentials bottleneck for the related studies’ models of Table 1. This can be
Execute Function (ReadAI) in Algorithm 1 achievable through reverting to the previously instantiated chaincode
end
and updating the embedded models off-chain based on their perfor­
else if Id ∈ KAID AND Fcn = QueryAI then
Inherits Credentials
mance, the amendment of the existing environment or the new on-chain
Execute Function (QueryAI) in Algorithm 1 data.
end On the other side, many researchers have provided recent studies
else with valuable contributions regarding the blockchain features and ap­
Revert Inputs and Show an error
plications in construction management. For instance, Ar et al. (2020)
end
provided a model to assist the decision-makers in evaluating the feasi­
bility of implementing blockchain in logistics operations. Kiu et al.
(2020) investigated the potential areas for utilizing blockchain in the
5. System testing construction industry. Tezel et al. (2020) also investigated the issues
associated with using the blockchain for construction supply chains,
The system is tested to assess its performance based on the latency of then provided a conceptual framework to tackle these issues. Elghaish
writing and reading, the addition of new participants, and the submis­ et al. (2020) developed a hyperledger fabric application to automati­
sion of incomplete data as guided by Almadhoun et al. (2018); Hasan cally execute and process the financial transactions of the ongoing
and Salah (2018); Xu et al. (2019). The writing latency for BBNs refers to construction projects. Lee et al. (2021) proposed a framework that in­
the inclusion time taken by a transaction to be packaged into a block and tegrates blockchain with digital twin to exchange and manage infor­
stored on the blockchain network. By contrast, reading latency refers to mation among the construction project stakeholders. Sheng, Ding, et al.
the time spent to retrieve a specific block’s data from the BBNs. For the (2020) introduced a novel blockchain-based solution to decentralize and
proposed system, the writing latency is less than or equal to 3000 ms, manage the quality information in construction projects.
while the reading latency is less than or equal to 500 ms. Both times Despite these contributions, there are two notable limitations; the
were calculated using IBM Cloud Function monitoring tools based on 50 studies are in the exploration and conceptualization phases, or the

7
K. Adel et al. Expert Systems With Applications 194 (2022) 116548

Fig. 5. Writing Process – Invocation of InferAI Function.

Fig. 6. Reading Process – Invocation of QueryAI Function.

studies present blockchain as a transaction-oriented technology only. consultations, recurring meetings, and administrative workloads. It also
This research takes a step forward from the phase of discussing and ensures the immutability and traceability of on-chain analyzed records
evaluating the theoretical aspects and the feasibility of blockchain in that increase the confidence and awareness about the reached decisions
construction to the design and implementation phases. It focuses on or analytics.
employing the BBN, not only as a decentralized record medium for The system design has a couple of limitations. First, the system is
exchanging raw textual and numeric data but also as a computing- limited to exchange and analyze textual and numeric data only. As a
oriented technology to simultaneously process and interpret trans­ direct result, the image processing models and visuals’ storage are not
actional data by embedding mathematical algorithms and ML models in guaranteed. However, this approach lowers the rate of increasing
the chaincodes. Moreover, the study’s system provides higher trans­ network storage “storage bloating problem”, the computational re­
parency and flexibility for the BBN users to agree upon which infor­ sources’ waste, and the high demand for internet bandwidth as guided
mation or decisions should be on-chain while diminishing face-to-face by Tanwar et al. (2019). Second, regarding the data feeding, repetitive

8
K. Adel et al. Expert Systems With Applications 194 (2022) 116548

or ambiguous data may be proposed or submitted, which could congest Almadhoun, R., Kadadha, M., Alhemeiri, M., Alshehhi, M., & Salah, K. (2018, 28 Oct.-1
Nov. 2018). A user authentication scheme of IoT devices using blockchain-enabled
the system’s ledgers, discompose the data’s authenticity, and corrupt the
fog nodes. Proceeding of 15th international conference on computer systems and
models’ future evolution. This can be tackled through the BBN’s vali­ applications (AICCSA), Aqaba, Jordan. doi: 10.1109/AICCSA.2018.8612856.
dation policy that allows the key actors to audit and evaluate each fed Ar, I. M., Erol, I., Peker, I., Ozdemir, A. I., Medeni, T. D., & Medeni, I. T. (2020).
input and computed outcome before being chained to avoid any possible Evaluating the feasibility of blockchain in logistics operations: A decision
framework. Expert Systems with Applications, 158, Article 113543. https://doi.org/
source of bias or error. 10.1016/j.eswa.2020.113543
Bagchi, S., Chakraborty, M., & Chattopadhyay, A. K. (2019, 30 November 2019).
7. Conclusion APDRChain: ANN Based Predictive Analysis of Diseases and Report Sharing Through
Blockchain. Proceedings of International Ethical Hacking Conference 2019,
Singapore. doi: 10.1007/978-981-15-0361-0_8.
Building an AI system within blockchain networks is a novel step Baker, H., Hallowell, M. R., & Tixier, A.-J.-P. (2020). AI-based prediction of independent
towards enabling trusted decentralized artificial intelligence for the construction safety outcomes from universal attributes. Automation in Construction,
118, Article 103146. https://doi.org/10.1016/j.autcon.2020.103146
construction industry. In this study, a configurable system for data Bamakan, S. M. H., Motavali, A., & Babaei Bondarti, A. (2020). A survey of blockchain
collection and decentralized inference has been introduced to serve as a consensus algorithms performance evaluation criteria. Expert Systems with
generic blockchain AI-models repository. The system is designed to Applications, 154, Article 113385. https://doi.org/10.1016/j.eswa.2020.113385
Cao, Y., Ashuri, B., & Baek, M. (2018). Prediction of unit price bids of resurfacing
guarantee confidential data sharing, AI decisions’ verification, and AI highway projects through ensemble machine learning. Journal of Computing in Civil
models’ upgrading and ownership traceability. The proposed system has Engineering, 32(5), 04018043. https://doi.org/10.1061/(ASCE)CP.1943-
been implemented via a four-step methodology while aiming to estimate 5487.0000788
Cerić, A. (2019, 2019). Blockchain strategy for minimizing information asymmetry in
the construction cost of road projects to support the construction firms in
construction projects. Proceedings of 14th International Conference: Organization,
bidding and financing alternatives decisions. The first step is building a Technology and Management in Construction, Zagreb, Croatia.
ML model as a multi-buffer perceptron artificial neural network while Chang, S. E., Chen, Y.-C., & Lu, M.-F. (2019). Supply chain re-engineering using
blockchain technology: A case of smart contract based tracking process.
employing the Levenberg Marquardt algorithm as a supervised dynamic
Technological Forecasting and Social Change, 144, 1–11. https://doi.org/10.1016/j.
nonlinear learning rule using NeuroSolution Software Version 5.92. The techfore.2019.03.015
second step is formulating the model in a chaincode template using Cheng, M.-Y., Kusoemo, D., & Gosno, R. A. (2020). Text mining-based construction site
JavaScript language. The third step is configuring the blockchain accident classification using hybrid supervised machine learning. Automation in
Construction, 118, Article 103265. https://doi.org/10.1016/j.autcon.2020.103265
network using IBM BaaS. The fourth step is developing the frontend Cho, H.-G., Kim, K.-G., Kim, J.-Y., & Kim, G.-H. (2013). A comparison of construction
interface that allows access and interaction with the blockchain network cost Estimation using multiple regression analysis and neural network in elementary
using IBM Cloud Functions. Further, the system performance has been school projects. Journal of the Korea Institute of Building Construction, 13(1), 66–74.
https://doi.org/10.5345/JKIBC.2013.13.1.066
evaluated based on the computing latency through 50 successive in­ Choi, J., Gu, B., Chin, S., & Lee, J.-S. (2020). Machine learning predictive model based on
vocations, the key actors scaling, and the submission of incomplete data. national data for fatal accidents of construction workers. Automation in Construction,
This research can be extended in the future to provide a completely 110, Article 102974. https://doi.org/10.1016/j.autcon.2019.102974
Chong, H.-Y., & Diamantopoulos, A. (2020). Integrating advanced technologies to uphold
decentralized AI platform that relies on BT and Inter-Planetary File security of payment: Data flow diagram. Automation in Construction, 114, Article
System to collect, refine, learn from the chained data and produce on- 103158. https://doi.org/10.1016/j.autcon.2020.103158
chain trained AI models. As such, it leverages the following features: Cirilovic, J., Vajdic, N., Mladenovic, G., & Queiroz, C. (2014). Developing cost estimation
models for road rehabilitation and reconstruction: Case study of projects in Europe
1) decentralizing the learning process instead of using conventional off-
and Central Asia. Journal of Construction Engineering Management, 140(3), 04013065.
chain training software packages; 2) upgrading the data acquisition, https://doi.org/10.1061/(ASCE)CO.1943-7862.0000817
circulation, and exchange to support image processing models and vi­ Corea, F. (2019). Applied artificial intelligence: Where AI can be used in business (Vol. 1).
Springer International Publishing.
suals’ storage; 3) avoiding the repetitive records using the content
Das, M., Luo, H., & Cheng, J. C. (2020). Securing interim payments in construction
hashing mechanism of Inter-Planetary File System. projects through a blockchain-based framework. Automation in Construction, 118,
Article 103284. https://doi.org/10.1016/j.autcon.2020.103284
Das, M., Tao, X., & Cheng, J. C. (2020, 2020). A Secure and distributed construction
CRediT authorship contribution statement
document management system using blockchain. Proceedings of the 18th
International Conference on Computing in Civil and Building Engineering, Cham.
Kareem Adel: Conceptualization, Methodology, Writing – original doi: 10.1007/978-3-030-51295-8_59.
draft. Ahmed Elhakeem: Supervision, Validation, Writing – review & Dinh, T. N., & Thai, M. T. J. C. (2018). AI and blockchain: A disruptive integration.
Computer, 51(9), 48–53. https://doi.org/10.1109/MC.2018.3620971
editing. Mohamed Marzouk: Supervision, Validation, Writing – review Ekramifard, A., Amintoosi, H., Seno, A. H., Dehghantanha, A., & Parizi, R. M. (2020). A
& editing. systematic literature review of integration of blockchain and artificial intelligence.
In Blockchain Cybersecurity, Trust and Privacy (pp. 147-160). Springer.
El-Kholy, A. (2019). Exploring the best ANN model based on four paradigms to predict
Declaration of Competing Interest delay and cost overrun percentages of highway projects. International Journal of
Construction Management, 1–19. https://doi.org/10.1080/15623599.2019.1580001
The authors declare that they have no known competing financial El-Kholy, A. M., Tahwia, A. M., & Elsayed, M. M. (2020). Prediction of simulated cost
contingency for steel reinforcement in building projects: ANN versus regression-
interests or personal relationships that could have appeared to influence based models. International Journal of Construction Management, 1–15. https://doi.
the work reported in this paper. org/10.1080/15623599.2020.1741492
Elbeltagi, E., Hosny, O., Abdel-Razek, R., & El-Fitory, A. (2014). Conceptual cost estimate
of libyan highway projects using artificial neural network. International Journal of
Appendix A. Supplementary data Engineering Research and Applications, 4(8), 56–66.
Elghaish, F., Abrishami, S., & Hosseini, M. R. (2020). Integrated project delivery with
Supplementary data to this article can be found online at https://doi. blockchain: An automated financial system. Automation in Construction, 114, Article
103182. https://doi.org/10.1016/j.autcon.2020.103182
org/10.1016/j.eswa.2022.116548. ElMousalami, H. H., Elyamany, A. H., & Ibrahim, A. H. (2018). Predicting conceptual
cost for field canal improvement projects. Journal of Construction Engineering
References Management, 144(11), 04018102. https://doi.org/10.1061/(ASCE)CO.1943-
7862.0001561
Fan, C.-L. (2020). Defect risk assessment using a hybrid machine learning method.
AACEI. (2020). Cost Estimate Classification System - As Applied in Engineering, Procurement,
Journal of Construction Engineering Management, 146(9), 04020102. https://doi.org/
And Contruction for The Process Industries. Retrieved (06/Sep/2020) from http://web.
10.1061/(ASCE)CO.1943-7862.0001897
aacei.org/docs/default-source/toc/toc_18r-97.pdf?sfvrsn=8.
Gondia, A., Siam, A., El-Dakhakhni, W., & Nassar, A. H. (2020). Machine learning
Adeli, H., & Wu, M. (1998). Regularization neural network for construction cost
algorithms for construction projects delay risk prediction. Journal of Construction
estimation. Journal of Construction Engineering and Management, 124(1), 18–24.
Engineering Management, 146(1), 04019085. https://doi.org/10.1061/(ASCE)
https://doi.org/10.1061/(ASCE)0733-9364(1998)124:1(18)
CO.1943-7862.0001736
Ahmadisheykhsarmast, S., & Sonmez, R. (2020). A smart contract system for security of
Gulati, P., Sharma, A., Bhasin, K., & Azad, C. (2020, May 14, 2020). Approaches of
payment of construction contracts. Automation in Construction, 120, Article 103401.
blockchain with AI: challenges & future direction. Proceedings of the International
https://doi.org/10.1016/j.autcon.2020.103401

9
K. Adel et al. Expert Systems With Applications 194 (2022) 116548

Conference on Innovative Computing & Communications (ICICC) 2020, New Delhi, International Journal of Construction Management, 1–14. https://doi.org/10.1080/
India. doi: 10.2139/ssrn.3600735. 15623599.2019.1692400
Günaydın, H. M., & Doğan, S. Z. (2004). A neural network approach for early cost Nassar, M., Salah, K., ur Rehman, M. H., & Svetinovic, D. (2020). Blockchain for
estimation of structural systems of buildings. International journal of project explainable and trustworthy artificial intelligence. Wiley Interdisciplinary Reviews:
management, 22(7), 595–602. https://doi.org/10.1016/j.ijproman.2004.04.002 Data Mining Knowledge Discovery, 10(1), e1340. doi: 10.1002/widm.1340.
Gunduz, M., Ugur, L. O., & Ozturk, E. (2011). Parametric cost estimation system for light Nawari, N. O., & Ravindran, S. (2019). Blockchain technologies in BIM workflow
rail transit and metro trackworks. Expert Systems with Applications, 38(3), environment. In Computing in Civil Engineering 2019: Visualization, Information
2873–2877. https://doi.org/10.1016/j.eswa.2010.08.080 Modeling, and Simulation (pp. 343-352). American Society of Civil Engineers Reston,
Guo, J., Wang, Q., Li, Y., & Liu, P. (2020). Façade defects classification from imbalanced VA.
dataset using meta learning-based convolutional neural network. Computer-Aided Pan, X., & Yang, T. (2020). Postdisaster image-based damage detection and repair cost
Civil Infrastructure Engineering, 35(12), 1403–1418. https://doi.org/10.1111/ estimation of reinforced concrete buildings using dual convolutional neural
mice.12578 networks. Computer-Aided Civil Infrastructure Engineering, 35(5), 495–510. https://
Gupta, I. (2020). Decentralization of Artificial Intelligence: Analyzing developments in doi.org/10.1111/mice.12549
decentralized learning and distributed AI networks. Panda, S. S., & Jena, D. (2020). Decentralizing AI using blockchain yechnology for secure
Harris, J. D., & Waggoner, B. (2019). Decentralized and collaborative AI on blockchain. decision making. In Advances in Machine Learning and Computational Intelligence (pp.
In 2019 IEEE International Conference on Blockchain (Blockchain). https://doi.org/ 687-694). Springer. doi: 10.1007/978-981-15-5243-4_65.
10.1109/Blockchain.2019.00057 Pandl, K. D., Thiebes, S., Schmidt-Kraepelin, M., & Sunyaev, A. (2020). On the
Hasan, H. R., & Salah, K. (2018). Proof of delivery of digital assets using blockchain and convergence of Artificial Intelligence and distributed ledger technology: A scoping
smart contracts. IEEE Access, 6, 65439–65448. https://doi.org/10.1109/ review and future research agenda. IEEE Access, 8, 57075–57095. https://doi.org/
ACCESS.2018.2876971 10.1109/ACCESS.2020.2981447
Hegazy, T., & Ayed, A. (1998). Neural network model for parametric cost estimation of Papakostas, N., Newell, A., & Hargaden, V. (2019). A novel paradigm for managing the
highway projects. Journal of Construction Engineering and Management, 124(3), product development process utilising blockchain technology principles. CIRP
210–218. https://doi.org/10.1061/(ASCE)0733-9364(1998)124:3(210) Annals, 68(1), 137–140. https://doi.org/10.1016/j.cirp.2019.04.039
Hegazy, T., Fazio, P., & Moselhi, O. (1994). Developing practical neural network Penzes, B., KirNup, A., Gage, C., Dravai, T., & Colmer, M. (2018). Blockchain technology in
applications using back-propagation [Article]. Microcomputers in civil engineering, 9 the construction industry: digital transformation for high productivity.
(2), 145–159. https://doi.org/10.1111/j.1467-8667.1994.tb00369.x Perera, S., Nanayakkara, S., Rodrigo, M., Senaratne, S., & Weinand, R. (2020).
Helo, P., & Shamsuzzoha, A. (2020). Real-time supply chain—A blockchain architecture Blockchain technology: Is it hype or real in the construction industry? Journal of
for project deliveries. Robotics Computer-Integrated Manufacturing, 63, Article Industrial Information Integration, 17, Article 100125. https://doi.org/10.1016/j.
101909. https://doi.org/10.1016/j.rcim.2019.101909 jii.2020.100125
Hijazi, A. A., Perera, S., Al-Ashwal, A. M., & Neves Calheiros, R. (2019). Enabling a single Rafiei, M. H., & Adeli, H. (2018). Novel machine-learning model for estimating
source of truth through BIM and blockchain integration. Proceedings of the 2019 construction costs considering economic variables and indexes. Journal of
International Conference on Innovation, Technology, Enterprise and Construction Engineering Management, 144(12), 04018106. https://doi.org/10.1061/
Entrepreneurship (ICITEE 2019), 24-25 November 2019, Kingdom of Bahrain. (ASCE)CO.1943-7862.0001570
Hunhevicz, J. J., & Hall, D. M. (2020). Do you need a blockchain in construction? Use Research-and-Markets-Store. (2020). Blockchain Market by Component (Platform and
case categories and decision framework for DLT design options. Advanced Engineering Services), Provider (Application, Middleware, and Infrastructure), Type (Private, Public,
Informatics, 45, Article 101094. https://doi.org/10.1016/j.aei.2020.101094 and Hybrid), Organization Size, Application Area (BFSI, Government, IT & Telecom), and
Inbaraj, X. A., & Chaitanya, T. R. (2020). Need to know about combined technologies of Region - Global Forecast to 2025. Retrieved 01/Jan/2021 from https://www.
blockchain and machine learning. In Handbook of Research on Blockchain Technology researchandmarkets.com/reports/5025113/blockchain-market-by-component-
(pp. 417–432). Elsevier. platform-and?w=4.
Kim, G.-H., Shin, J.-M., Kim, S., & Shin, Y. (2013). Comparison of school building Sakhakarmi, S., Park, J., & Cho, C. (2019). Enhanced machine learning classification
construction costs estimation methods using regression analysis, neural network, and accuracy for scaffolding safety using increased features. Journal of Construction
support vector machine Journal of Building Construction and Planning Research, 1(1), Engineering Management, 145(2), 04018133. https://doi.org/10.1061/(ASCE)
7, Article 29576. doi: 10.4236/jbcpr.2013.11001. CO.1943-7862.0001601
Kim, G.-H., Yoon, J.-E., An, S.-H., Cho, H.-H., & Kang, K.-I. (2004). Neural network model Salah, K., Rehman, M. H. U., Nizamuddin, N., & Al-Fuqaha, A. (2019). Blockchain for AI:
incorporating a genetic algorithm in estimating construction costs. Building Review and open research challenges. IEEE Access, 7, 10127–10149. https://doi.org/
Environment, 39(11), 1333–1340. https://doi.org/10.1016/j.buildenv.2004.03.009 10.1109/ACCESS.2018.2890507
Kim, G. H., Seo, D. S., & Kang, K. I. (2005). Hybrid models of neural networks and genetic Sanni-Anibire, M. O., Zin, R. M., & Olatunji, S. O. (2020). Machine learning model for
algorithms for predicting preliminary cost estimates. Journal of Computing in Civil delay risk assessment in tall building projects. International Journal of Construction
Engineering, 19(2), 208–211. https://doi.org/10.1061/(ASCE)0887-3801(2005)19:2 Management, 1–10. https://doi.org/10.1080/15623599.2020.1768326
(208) Sarpatwar, K., Vaculin, R., Min, H., Su, G., Heath, T., Ganapavarapu, G., & Dillenberger,
Kim, S.-K., & Huh, J.-H. (2020). Artificial neural network blockchain techniques for D. (2019). Towards enabling trusted artificial intelligence via blockchain. In Policy-
healthcare system: Focusing on the personal health records. Electronics, 9(5), 763. Based Autonomic Data Governance (pp. 137-153). Springer. doi: 10.1007/978-3-030-
https://doi.org/10.3390/electronics9050763 17277-0_8.
Kim, S., Park, H., & Lee, J. (2020). Word2vec-based latent semantic analysis (W2V-LSA) Sgantzos, K., & Grigg, I. (2019). Artificial intelligence implementations on the
for topic modeling: A study on blockchain technology trend analysis. Expert Systems blockchain. Use cases and future applications. Future Internet, 11(8), 170. https://
with Applications, 152, Article 113401. https://doi.org/10.1016/j.eswa.2020.113401 doi.org/10.3390/fi11080170
Kiu, M., Chia, F., & Wong, P. (2020). Exploring the potentials of blockchain application Sheng, D., Ding, L., Zhong, B., Love, P. E., Luo, H., & Chen, J. (2020). Construction
in construction industry: A systematic review. International Journal of Construction quality information management with blockchains. Automation in Construction, 120,
Management, 1–10. https://doi.org/10.1080/15623599.2020.1833436 Article 103373. https://doi.org/10.1016/j.autcon.2020.103373
Krittanawong, C., Rogers, A. J., Aydar, M., Choi, E., Johnson, K. W., Wang, Z., & Sheng, D., Luo, H., & Zhong, B. (2020, 28 June – 1 July 2020). Formal modeling of smart
Narayan, S. M. (2020). Integrating blockchain technology with artificial intelligence contracts for quality acceptance in construction. Creative Construction e-Conference
for cardiovascular medicine. Nature Reviews Cardiology, 17(1), 1–3. https://doi.org/ 2020, Opatija, Croatia. doi: 10.3311/CCC2020-026.
10.1038/s41569-019-0294-y Shi, F., Wang, Q., & Wang, Y. (2019). Research on top-level redesign of smart
Lee, D., Lee, S. H., Masoud, N., Krishnan, M. S., & Li, V. C. (2021). Integrated digital twin construction system based on case study. In ICCREM 2019: Innovative Construction
and blockchain framework to support accountable information sharing in Project Management and Construction Industrialization (pp. 117–124). VA: American
construction projects. Automation in Construction, 127, Article 103688. https://doi. Society of Civil Engineers Reston.
org/10.1016/j.autcon.2021.103688 Sreckovic, M., & Windsperger, J. (2019). Decentralized autonomous organizations and
Li, J., Greenwood, D., & Kassem, M. (2019). Blockchain in the built environment and network Design in AEC: a conceptual framework.
construction industry: A systematic review, conceptual models and practical use Szabo, N. (1994). Smart contracts. Retrieved 6 Sep 2020 from https://www.fon.hum.uva.
cases. Automation in Construction, 102, 288–307. https://doi.org/10.1016/j. nl/rob/Courses/InformationInSpeech/CDROM/Literature/LOTwinterschool2006/
autcon.2019.02.005 szabo.best.vwh.net/smart.contracts.html.
Li, J., Kassem, M., & Watson, R. (2020). A blockchain and smart contract-based Tang, H., Shi, Y., & Dong, P. (2019). Public blockchain evaluation using entropy and
framework to increase traceability of built assets. CIB W78 Information Technology TOPSIS. Expert Systems with Applications, 117, 204–210. https://doi.org/10.1016/j.
for Construction Conference São Paulo, Brazil. doi: 10.46421/2706-6568.37.2020. eswa.2018.09.048
paper025. Tanwar, S., Bhatia, Q., Patel, P., Kumari, A., Singh, P. K., & Hong, W.-C. (2019). Machine
Maeda, H., Kashiyama, T., Sekimoto, Y., Seto, T., & Omata, H. (2020). Generative learning adoption in blockchain-based smart applications: The challenges, and a way
adversarial network for road damage detection. Computer-Aided Civil Infrastructure forward. IEEE Access, 8, 474–488. https://doi.org/10.1109/ACCESS.2019.2961372
Engineering, 36, 47–60. https://doi.org/10.1111/mice.12561 Tezel, A., Papadonikolaki, E., Yitmen, I., & Hilletofth, P. (2020). Preparing construction
Mamoshina, P., Ojomoko, L., Yanovich, Y., Ostrovski, A., Botezatu, A., Prikhodko, P., … supply chains for blockchain technology: An investigation of its potential and future
Zhebrak, A. (2018). Converging blockchain and next-generation artificial directions. Frontiers of Engineering Management, 1–17. https://doi.org/10.1007/
intelligence technologies to decentralize and accelerate biomedical research and s42524-020-0110-8
healthcare. Oncotarget, 9(5), 5665–5690. https://doi.org/10.18632/ Vyas, S., Gupta, M., & Yadav, R. (2019, 4-6 Feb. 2019). Converging blockchain and
oncotarget.22345 machine learning for healthcare. 2019 Amity International Conference on Artificial
Matel, E., Vahdatikhaki, F., Hosseinyalamdary, S., Evers, T., & Voordijk, H. (2019). An Intelligence (AICAI), doi: 10.1109/AICAI.2019.8701230.
artificial neural network approach for cost estimation of engineering services. Wang, W. M., Guo, H., Li, Z., Shen, Y., & Barenji, A. V. (2018). Towards open and
automated customer service: A blockchain-based automl framework. Proceedings of

10
K. Adel et al. Expert Systems With Applications 194 (2022) 116548

the 2nd International Conference on Computer Science and Application Engineering, Yao, L., Dong, Q., Jiang, J., & Ni, F. (2020). Deep reinforcement learning for long-term
Hohhot China. doi: 10.1145/3207677.3277939. pavement maintenance planning. Computer-Aided Civil Infrastructure Engineering, 35,
Wang, Z., Wang, T., Hu, H., Gong, J., Ren, X., & Xiao, Q. J. A.i. C. (2020). Blockchain- 1230–1245. https://doi.org/10.1111/mice.12558
based framework for improving supply chain traceability and information sharing in Ye, X., & König, M. (2020, July 14, 2020). Framework for automated billing in the
precast construction. Automation in Construction, 111, Article 103063. https://doi. construction industry using BIM and smart contracts. Proceedings of the 18th
org/10.1016/j.autcon.2019.103063 International Conference on Computing in Civil and Building Engineering. ICCCBE
Xu, X., Lu, Q., Liu, Y., Zhu, L., Yao, H., & Vasilakos, A. V. (2019). Designing blockchain- 2020., Cham. doi: 10.1007/978-3-030-51295-8_57.
based applications a case study for imported product traceability. Future Generation Zhang, Z., Yuan, Z., Ni, G., Lin, H., & Lu, Y. (2020). The quality traceability system for
Computer Systems, 92, 399–406. https://doi.org/10.1016/j.future.2018.10.010 prefabricated buildings using blockchain: An integrated framework. Frontiers of
Xue, F., & Lu, W. (2020). A semantic differential transaction approach to minimizing Engineering Management, 1–19. https://doi.org/10.1007/s42524-020-0127-z
information redundancy for BIM and blockchain integration. Automation in Zhong, B., Wu, H., Ding, L., Luo, H., Luo, Y., & Pan, X. (2020). Hyperledger fabric-based
Construction, 118, Article 103270. https://doi.org/10.1016/j.autcon.2020.103270 consortium blockchain for construction quality information management. Frontiers
Yang, R., Wakefield, R., Lyu, S., Jayasuriya, S., Han, F., Yi, X., … Chen, S. (2020). Public of Engineering Management, 1–16. https://doi.org/10.1007/s42524-020-0128-y
and private blockchain in construction business process and information integration. Zhu, M., Zhu, H., Guo, F., Chen, X., & Ju, J. W. (2020). Tunnel condition assessment via
Automation in Construction, 118, Article 103276. https://doi.org/10.1016/j. cloud model-based random forests and self-training approach. Computer-Aided Civil
autcon.2020.103276 Infrastructure Engineering. https://doi.org/10.1111/mice.12601

11