Evolution of Smart Home Energy Management System Using

Internet of Things and Machine Learning Algorithms

Tarana Singh1, Arun Solanki2 and Sanjay Kumar Sharma3
1
Gautam Buddha University, Department of Computer Science and Engineering, School of Information and
Communication Technology, Greater Noida, India.
2
Gautam Buddha University, Department of Computer Science and Engineering, School of Information and
Communication Technology, Greater Noida, India.
3
Gautam Buddha University, Department of Computer Science and Engineering, School of Information and
Communication Technology, Greater Noida, India.

Abstract
The cost and demand of energy are increasing day by day, leading the domain to find new and
smart ways to monitor, control, and save energy. In the smart city, smart energy management
systems help to resolve the energy management problem. Smart energy management systems cut
the cost of energy in smart houses or buildings with their recommendations and predictions. This
paper proposed a 5-layer architecture for a Home Energy Management System (HEMS), which
collects real-time data; analyzes the patterns from the data, and further feed the patterns into the
recommendation system to generate recommendations to save energy. A massive amount of data
is collected using different sensors in the proposed architecture. This architecture has different
layers, all of which are dedicated to performing specific tasks accordingly. The different
preprocessing and Machine Learning (ML) techniques like Simple Linear Regression (SLR),
Decision Tree Regression (DTR), Random Forest Regression (RFR), K-Nearest Neighbor
Regression (KNNR), and Support Vector Regression (SVR) are used for data analysis. This study
finds that Decision Tree regression (0.9999) and Random Forest Regression (0.9999) achieved
good scores compared to the Simple Linear Regression (0.9901), K-Nearest Neighbor (0.9720),
and Support Vector Regression (0.9966). The values of Root Mean Square Error (RMSE) and
Mean Absolute Error (MAE) for SLR (0.9900 and 0.0099), DTR (0.0439 and 0.0019), RFR
(0.0427 and 0.0018), KNNR (0.0285 and 0.1690) and SVR (0.1394 and 0.0194). Thus, this paper
concludes that the decision tree and random forest regression are having high scores and less error
in comparison to the other algorithms. These regression techniques for data analysis can be used
for the recommendation of energy management using the proposed architecture.

Key Words
Smart City, Energy management, Machine Learning, Internet of Things, Big Data.

1. INTRODUCTION

The effective use of energy in smart homes saves money, improves sustainability, and decreases the
carbon impact on a wide scale [1, 2, 3]. The vast volume of data generated throughout a nation poses several
data storage, management, and analysis issues. IoT and Big Data are logical solutions to these problems [4,
5]. IoT technologies may offer a ubiquitous computing platform for sensing, monitoring, and controlling
the energy use of home appliances on a wide scale. This information is gathered utilizing a variety of
wireless sensors put in residential units [6, 38]. Big Data technology may be used to collect simultaneously
and analyze the data. Data is gathered, evaluated, and translated into usable information in reports, graphs,
and charts that utilize predictive analytics and sophisticated technologies [7, 8, 9].
HEMS is a critical component of the smart grid environment because it enables load management among
homeowners to reduce energy costs while flexibly supporting high penetration Renewable Energy Sources
ACI’22: Workshop on Advances in Computation Intelligence, its Concepts & Applications at ISIC 2022, May 17-19, Savannah, United States
EMAIL: [email protected] (A. 1); [email protected] (A. 2); [email protected] (A. 3)
ORCID: 0000-0002-9035-0068 (A. 1); 0000-0002-4359-406X (A. 2); 0000-0001-7918-4938 (A. 3)
©️ 2020 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)

135
(RES) at both the distant and local levels. Traditional household appliances, as well as developing ones
such as Energy Storage Systems (ESS), Electrical Vehicles (EV) [11, 12], and others, must be considered
in an efficient and cost-effective HEMS. The new appliances allow the HEMS to reduce costs, even more,
minimize peak pressures, and overcome the volatility of RES production [13, 39, 40]. Unmanaged EV
charging and discharging, for example, might increase the peak demand, lead to dangerous overload, and
damage local distribution lines. To reduce the user’s power cost and discontent, a convex programming
home energy optimization framework including schedule-based appliances, battery-assisted appliances,
and model-based appliances is proposed in a few research articles [14, 15]. There are several writers that
have worked on Home Automation (HA) and energy conservation. In [45] a smart home energy
management architectural proposal, the authors recommended that the sensors can be mounted for the
detection and measurement of environmental data in their study. The microcontroller, which is the
fundamental component of the design, receives the gathered data. The microcontroller is connected to
various components of the architecture using various technologies such as Zigbee, X-bee shield, and so on.
Internet connection, interpretation, and processing of data from various sensors transmit control signals to
the architecture's appliances or actuators. It also delivers real-time environmental information to the website
and handles requests sent from distant users via the webserver, according to the author. [47] suggested a
home automation architecture that is part of the IoT application area and provides several potentials for
building new beneficial applications. The author said that home automation (HA) is a collection of
approaches for automating a house that incorporates technology into security, energy management, and
welfare. Comfort is an important aspect of a home automation system since it encompasses all of the
measures taken to enhance how occupants feel in their homes. [48] proposed a self-learning
SHEMS architecture. The author said that their suggested system optimizes residential household units
based on consumer comfort and energy cost, as well as reducing power supply overloading. For decision-
making, the author employed a demand-side management system, a supply-side management system, price
forecasting, price clustering, energy warning systems, and so on. The suggested system was verified by the
author gathering real-time power usage data in Singapore. The author was doing a real-time case study. The
structure of the smart home energy management system was addressed in [49]. The smart controller, which
offers system management features such as logging, monitoring, and control, was mentioned by the author
as a component of HEM. The smart microcontroller, according to the author, gathers real-time power
consumption data from programmable and non-programmable appliances in order to apply effective
demand management techniques.
Smart house IoT appliances [42, 43], Big Data [39], and Machine Learning Algorithms [37, 39] have
limited capacities. As a result, more data handling choices must be included to properly gather, handle, and
analyze massive amounts of information. Massive volumes of data are being collected and analyzed using
big data analytics tools. researchers also allow for the appropriate interpretation and use of large volumes
of sensor data. ML, on the other hand, is an artificial intelligence subfield that analyses algorithms and
statistical models based on patterns and inferences that systems use to accomplish their goals. ML is also
often used in real-time applications because to its viability and endurance. Simultaneously, ML addresses
learning-related problems and discovers the background and characteristics of such difficulties in order to
learn from them and enhance system performance. Finally, ML is classified into reinforcement,
unsupervised, and supervised learning, is used to carry out tasks that need previously obtained data. The
IoT paradigm, machine learning, big data technologies, and the application of these technologies in real-
time are now the problems in the smart home space. The suggested architecture for energy management in
smart homes utilizes IoT [40, 41], Bigdata [38, 39], and Machine Learning [39]. This study will assist in
overcoming the obstacles that have previously been encountered. The proposed architecture enables real-
time data collection, with the acquired information being saved in a Google document. This information is
used for data analysis. This study uses five machine learning techniques to analyze the data i.e., SLR, DTR,
RFR, KNNR, and SVR. The RMSE and MAE for the algorithms is also determined to validate these results.
The paper is divided into eight sections. The introduction to the domain is covered in the first section of
the paper. The review of the related area is described in the second section of the paper. The proposed
architecture's process flow is discussed in the third section of the paper. The fourth section of the paper

136
discusses the proposed architecture of the energy management system. The data collection and
preprocessing techniques are being discussed in the fifth section of the paper. The sixth section presents the
evolution and analysis of the proposed system. The seventh section of the paper presents the discussion and
visual representation of the results. The conclusion with future work is discussed in the eighth section of
the paper, followed by the references.

2. Related Work

The smart city is a highly trendy topic these days, and it's drawing a lot of academics and experts to
work in its many subdomains. There are many subdomains of smart cities, such as transportation, health
care, education, agriculture, and so on, with energy management at the top of the list. Energy management
is becoming a more substantial area to deal with these days. Energy is a vital necessity worldwide, whether
for healthcare, transportation, education, or agriculture. Energy conservation is becoming a problem in
every discipline at the same time. Many researchers have worked and continue to try to solve this challenge.
Here are some of the research studies undertaken by various researchers to address these issues.
Sivaoragash et al. [16] focused on location-based optimum service selection for data management in
smart grids using cloud computing. The author concentrated on the effective use of energy. According to
the author, this is only achievable with the support of contemporary information technology. The author of
this study offered a user-aware power regulation model for smart grids, as well as a location-based service
selection strategy. The author then mentioned that they would also be introducing a secure and reliable
solution. Liang et al. [17] discussed deep reinforcement learning for smart home energy control. The author
analyzed the energy cost reduction challenge for smart homes in the absence of a building thermal dynamics
model while keeping a pleasant temperature range in mind. The suggested Model's simulation results, based
on real-world traces, indicate the algorithm's efficacy and unpredictability. Zafar et al. [18] proposed a study
on the home energy management system and emphasized its principles, setup, and smart grid technologies.
The author provided a thorough overview of the literature on home energy management systems, including
references to key ideas, configurations, and supporting technologies. Dinh et al. [19] discussed a home
energy management system that uses renewable energy and energy storage in the main grid and sells power.
The author of this study developed an architecture with RES and ESS that considers the use of power from
the main grid and electricity selling. Compared to earlier studies, the author can reduce residential energy
consumption by 19.7% using the offered strategies. Designing, developing, and deploying an IoT-based
smart energy management system was explored by Saleem et al. [44]. The author studied and examined
several presentations, implementation, and validation of an IoT-based smart energy management approach,
as well as the advantages of overcoming consumer energy management difficulties. To interact with any
software-based smart solution, the author uses a variety of communication interfaces and protocols. For
data collection and analysis, the author used Entrack software. The author validated the system in four
distinct buildings. The case study analysis of this work demonstrates the system's efficiency.
Li et al. [48] proposed a HEMS, DSMS, SSMS unified for real-time operations of a smart home. The
author discussed that the unified structure had some abilities such as worth prediction, value cluster, and
energy aware structure which helped in enhancing its roles, it was done using ML techniques. The author
discussed the proposed work with the self-energy HEMS infrastructure. Zekić-Sušac et al. [51] proposed a
HEMS-IoT system which explores big data, and ML for household ease, protection, and saving energy.
The author utilized the J48 ML model and Weka API to acquire consumer behavior and energy utilization
outlines and categorize households with the respect to energy utilization. The author discussed their work
with the help of the proposed architecture in the paper. Chouaib et al. [45] proposed a preposition of a
general microcontroller-based HEMS architecture. The author explained that the proposed system aims to
reduce the power utilization in smart homes, which was obtained by monitoring and regulating electrical
home appliances. The author’s anticipated system has been used to proposal and apply an effective lighting
system to reduce energy utilization in the smart home.

137
3. Process Flow Diagram

Figure 1 represents the process flow of the proposed architecture. The proposed architecture of HEMS
collects real-time data. The collected data is being used for further analysis. The patterns of energy
consumption are being analyzed using ML techniques in the administration layers of the proposed
architecture. These observed patterns are further fed into the recommendation system in the service layer.

Figure 1: Process flow diagram of the proposed work

Following steps are carried out for proposed work.
Step 1: Data is collected in the device layer using different devices like sensors, controllers, actuators,
gateways, etc.
Step 2: Preprocessing of the collected data is being done using different ML techniques.
Step 3: Using the preprocessed data obtained in step 2, the patterns of the energy consumption are
obtained and different parameters which affect the energy consumption patterns are also analyzed.
Step 4: The patterns obtained in step 3 are fed into the recommendation system to generate the
recommendations which can help to save energy.
Step 5: With the help of a pattern finder and the recommendation system in steps 3 and 4, energy-saving
recommendations are being generated which helps in saving energy.

4. Proposed Architecture of Home Energy Management System

This paper proposed a new architecture for HEMS as shown in figure 2. This architecture has five layers,
and each layer is dedicated to performing some specific tasks. The first layer of the architecture is the device
layer dedicated to data collection from the smart home environment. The second is for the communication
and information layer, which contains the information about the user, devices of the smart home, and the
sensors. The third layer is the administration layer, which deals with the management of the information
contained in the system with the pattern finder system. The fourth layer is the service and security layer
dedicated to providing the services and security. The primary component of this layer is the
recommendation system that utilizes the pattern data obtained from the previous layer and used those
patterns for the recommendations. This layer is also having the security layer that performs the
authentication and authorization. Few authors [45, 46, 47, 48, 49,50, 51] proposed the architecture of smart
home energy management systems. The user interfaces layer at the end is interfaced with the users in real-
time.

138
Figure 2: Layered Architecture of Energy Management system in Smart Home

4.1 Device Layer

This layer is divided into two sub-modules i.e., Smart home which consists of different types of devices
in the smart home, and Data collection which contains different data collection devices.

4.1.1 Smart Home Module

HEMS is used to optimize the operating time of household appliances. HEMS makes the consumer's
life easier by lowering electricity degeneracy and expense. Appliances are divided into three groups for
optimization: shiftable, non-shiftable, and fixed appliances. Figure 3 depicts the classification of different
appliances used in the smart home environment

Figure 3: Types of Appliances used in Smart Homes or Buildings

4.1.2 Data Collection Module

IoT component mentioned in figure 4 i.e., gateways, sensors, actuators, and controllers permits
data collection from numerous household appliances. The device layer also controls actuators and home
automation devices.

139
 Figure 4: Devices used for Data Collection in Smart Homes

4.2 Network and Information Layer

This layer is divided into two sub-modules i.e., network module and the information module.

4.2.1 Network Module

To choose the communication protocols for each home device, this layer evaluates components
such as a collection of sensors, HTTP and TCP/IP, Bluetooth, WiFi, and 4G communication. The
communication layer allows other levels in the architecture to communicate with one another. The HEMS
communication layer utilizes Zigbee, TCP/IP, HTTP/IP, Bluetooth, WiFi, and other protocols/technologies.

4.2.2 Information Module

The information produced in the device layer is saved in this layer. The information layer, in
particular, uses modules to handle five categories of data: user profile, device profile, sensor data, service
data, and pattern data for the energy management system. The pattern data module monitors the data felt
and gathered by the various smart home sensors using these five modules. This information is examined to
find the best patterns in the data, which are then sent into the recommendation system.

4.3 Administration Layer

This layer executes and controls the tasks necessary to satisfy the application layer's user needs.
The service layer ensures communication between the presentation and administration layers using the
REST API, Recommendation system, and service selector. User management, Home and Device
Management, Optimized Pattern Finder System, and Dashboard are the four categories of tasks done by the
administration layer.

4.4 Service and Security Layer

This layer is separated into two sub-modules i.e., Service Layer and Security Layer.

4.4.1 Service Module

The application and management layer are linked by this layer. Additionally, this layer includes
several REST APIs that enable users to access all HEMS features. IoT Service Selector, Recommendation
System, and REST APIs are the primary components of this layer.

4.4.2 Security Module

140
 This layer ensures data security and, as a result, the security and privacy of the gathered data from
the device layer. This layer covers two security components: authorization (API Key, Basic Authorization,
Hash Message Auth. Code, OAuth, and so on) and authentication (API Key, Basic Authorization, Hash-
Based Message Authorization Code, OAuth, and so on) (Pass Authentication Protocol, THAP, EAP, SFA,
TFA, MA, etc.).

4.5 User Interface Layer

This layer establishes a connection among the operator and the structure over a mobile application or
web application.
Thus the proposed architecture has five layers which make system maintenance easier and increase
scalability. The architecture aims to create an optimal pattern finder system that analyzes energy
consumption patterns in the home environment. These discovered patterns are used in the service layer of
recommendation systems. These recommendation systems provide customers with personalized
recommendations and aid in reducing energy use.

5. Data Collection and Preprocessing:

The real-time data is collected using different sensors like temperature, humidity, pressure, gas, fire,
laser object detection, voltage, electricity, etc. The data contains different parameters that affect the energy
requirement in the smart homes’ environment. A vast data set is required to implement the machine learning
algorithm. The steps of data collection using the proposed framework are represented in figure 5(B). The
design and application of the proposed system using the new architecture are shown in figure 5(A).

Step 1: In this IoT framework, the sensors are connected to the Arduino Uno.
Step 2: Arduino Uno is interfaced with NodeMCU and reads the input from sensors like fire, temperature,
humidity, gas, voltage, and electricity sensors.
Step 3: NodeMCU collects the data and stores it in the Google Spreadsheet.

Fig 5(A) Fig 5(B)

Figure 5(A, B): Framework and Steps of Data Collection Process using the framework

Using the proposed system, a chunk of data is collected which is not feasible for ML implementation.
So, this study used the data set available on online repositories [25] to implement and analyze the existing
ML techniques on the proposed model. Using the proposed system, the sample of the collected dataset is
represented in figure 6(A, B). The values obtained from different sensors are represented in figure 6 (A)
with the parameters i.e., Date, Time, Voltage, Current, Gas. Figure 6 (B) represents the complete
information of the collected dataset i.e., the data of the parameters, the total number of entries, the total
number of rows and columns, etc.

141
 Fig. 6(A) Fig. 6 (B)
Figure 6(A, B):): Representation of the samples of the data collected using IoT

The preprocessing of the dataset is being done by removing all the missing and null values, and
obtaining a feasible dataset for feeding into the ML models. The parameters are divided into the feature
matrix and the prediction vector. The feature matrix (independent variables) has eight parameters:
temperature, humidity, wind speed, visibility, pressure, summary, and precip type; on the other hand, the
prediction vector (dependent variable) has one parameter that is apparent temperature. Here the ML model
is being trained with the help of the feature matrix to predict the evident temperature according to the given
parameters of the feature matrix.
The dataset is separated into two parts i.e. training and testing sets in an 80-20% ratio, i.e., 80% of
the dataset is the training set, and 20% of the data is the testing set. The different machine learning
algorithms i.e. SLR, DTR, RFR, KNNR, SVR are being implemented. The ML models used in the proposed
work are discussed in section 6.

6. Analysis of Machine Learning Techniques

Machine Learning is becoming the most powerful tool to face the challenges in technological
development in the smart city. In this section, SLR, DTR, RFR, KNNR, SVR are implemented and analyzed
for the recommendation of energy management in HEMS.

6.1 Simple Linear Regression

By fitting a line to the observed data, this Model estimates the connection between the variables.
It's used to determine the asset of the link between two variables and the values of the dependent variables
at a given value of the independent variable [21, 22, 23]. This ML model is implemented on the dataset to
predict the apparent temperature. The score achieved by this model is 0.9901 given in table 1. Graph 1 is
representing the actual and predicted values using the SLR model.

Graph 1. Representation of Actual and Predicted values using Simple Linear Regression

6.2 Decision Tree Regression

142
 A decision tree constructs regression or classification models in the shape of a tree structure. It
gradually cuts down a dataset into smaller and smaller sections while also developing an associated
decision tree. A tree containing decision nodes and leaf nodes is the result. Each branch of a decision
node represents a value for the property being examined. A decision on the numerical objective is
represented by a leaf node. The root node is the highest decision node in a tree that corresponds to the
best predictor. Both category and numerical data may be handled by decision trees [24, 25]. This ML
model is implemented on the dataset to predict the apparent temperature. The score achieved by this
model is 0.9999 given in table 1. Graph 2 is representing the actual and predicted values using the DTR
model.

Graph 2. Representation of Actual and Predicted values using Decision Tree Regression

6.3 Random Forest Regression

At training, random forests (RF) create many individual decision trees. To create the final forecast,
the mode of the classes for classification or the mean prediction for regression, the predictions from all
trees are combined. Ensemble approaches are named because they conclude based on a group of
outcomes. With the use of several decision trees and a method called Bootstrap and Aggregation, often
known as bagging, it can complete both regression and classification tasks. Instead, depending on
individual decision trees, the main concept is to aggregate numerous decision trees to determine the
outcome [26]. This ML model is implemented on the dataset to predict the apparent temperature. The
score achieved by this model is 0.9999 given in table 1. Graph 3 is representing the actual and predicted
values using the RFR model.

Graph 3. Illustration of Actual and Predicted values using Random Forest Regression

6.4 K-Nearest Neighbor Regression

Both “classification and regression” both the issues may be solved with the KNN technique. The
KNN forecasts the values of novel data points based on 'feature similarity.' This implies that a value is
given to the new point depending on its similarity to the points in the training set. The distance between
the new point and each training point must first be calculated. There are many ways of determining this
distance, the most well-known of which are the Euclidian, Manhattan (continuous), and Hamming

143
 distances (for categorical) [27, 28]. This ML model is implemented on the dataset to predict the apparent
temperature. The score achieved by this model is 0.9720 given in table 1. Graph 4 is representing the
actual and predicted values using the KNNR model.

Graph 4. Illustration of Actual and Predicted values using KNNR

6.5 Support Vector Machine Regression

Kernel, Hyperplane, and verdict boundaries are three crucial factors in this machine learning model.
Kernel supports in the exploration for a hyper-plane in a high-dimensional space while dropping the
computation rate. As the size of the data grows, the computing cost will rise. A hyperplane splits the line
between two data classes in a support vector machine. This line is used to forecast the constant output in
SVR. The verdict boundary may be viewed as a separation line for simplification, with constructive
instances on one side and undesirable instances on the other. The examples on this line may be considered
as whichever constructive or undesirable [28]. This ML model is implemented on the dataset to predict
the apparent temperature. The score achieved by this model is 0.9966 given in table 1. Graph 5 is
representing the actual and predicted values using the SVR model.

Graph 5. Representation of Actual and Predicted values using Support Vector Regression

It is detected that the DTR and RFR can predict the apparent temperature with a high score i.e.,
0.9999 (for both the models). The score for the rest of the ML models is 0.9720 (for KNNR), 0.9966 (for
SVR), and 0.9901 (for SLR), which is less than the DTR and RFR scores. The RMSE and MAE values are
also higher than the SLR, DTR, RFR, KNNR, and SVR are discussed in section 7.

7. Discussion and Visual Representation of the Results

The ML techniques i.e. SLR, DTR, RFR, KNNR, SVR discussed in section 6 used to analyze the data. The
RMSE and MAE [31, 32, 33] values validate the prediction in comparison to actual values. The ML model
has considered the best model for problem-solving if it has a high score and a low error value.

7.1 Root Mean Square Error

144
 It's also known as root mean square deviation, most typically used to assess the accuracy of
predictions. The standard deviation of the errors that occur while predicting a dataset is the RMSE given in
equation 1. It uses Euclidean Distance to demonstrate how forecasts differ from actual values [34]. The
RMSE values obtained by implementing the SLR, DTR, RFR, KNNR, and SVR are given in table 1.
Let N is the number of data points in the dataset, 𝑦𝑖 is the ith measurements of the datapoint from the
whole dataset, 𝑦𝑝𝑟𝑒𝑑(𝑖) is the predictions for the ith measurement in the whole dataset. RMSE is the standard
deviation of the forecast errors. It is being generally calculated to verify the experimental results of the
regression analysis. The following steps are being carried out to calculate the RMSE.
Step 1: For all the predicted values, calculate the difference from the corresponding actual value, i.e., given
in equation (a);
(𝒚𝒊 − 𝒚𝒑𝒓𝒆𝒅(𝒊) )
(a)

Step 2: Square the differences obtained in step 1 i.e., given in equation (b);
𝟐
||(𝒚𝒊 − 𝒚𝒑𝒓𝒆𝒅(𝒊) )||
(b)

Step 3: Sum all the “squared differences” calculated in step 2, I.e., given in equation (c).
𝒌 (c)
𝟐
∑ ||(𝒚𝒊 − 𝒚𝒑𝒓𝒆𝒅(𝒊))||
𝒊=𝟏
Step 4: Calculate the average of the “sum of squared differences” derived in Step 3, which is called MSE
or Mean Squared Error, i.e., given in equation (d).
𝒌
𝟐 (d)
∑ ||(𝒚𝒊 − 𝒚𝒑𝒓𝒆𝒅(𝒊) )|| /𝑵
𝒊=𝟏
Step 5: Finally take the square root of the values derived in step 4, which is RMSE, i.e., given in equation
(1).

𝒌
𝟐
RMSE = √∑ ||(𝒚𝒊 − 𝒚𝒑𝒓𝒆𝒅(𝒊) )|| /𝑵 (1)
𝒊=𝟏

7.2 Mean Absolute Error

The error is the absolute difference between the real or actual values and the expected values. If the
findings have a negative sign, it is ignored by the absolute difference.
MAE = Actual Values – Predicted Values (2)
This function computes the average of this error over all samples in a dataset and returns the result
given in equation 2. But, in certain cases, this value may not be the most important factor to consider when
dealing with a real-life issue since the data that is being used to construct and assess the Model is the same,
implying that the Model has never been exposed to genuine [36]. The MAE values obtained by

145
implementing the SLR, DTR, RFR, KNNR, and SVR are given in table 1. As a result, it may perform well
on visible data but may struggle or fail terribly when confronted with actual, hidden data.
Table 1:
Result Analysis of the different machine learning algorithms
Machine Learning Algorithm Score RMSE MAE
Simple Linear Regression (SLR) 0.9901 0.9900 0.0099
Decision Tree Regression (DTR) 0.9999 0.0439 0.0019
Random Forest Regression (RFR) 0.9999 0.0427 0.0018
K-Nearest Neighbor Regression (KNNR) 0.9720 0.0285 0.1690
Support Vector Machine Regression (SVR) 0.9966 0.1394 0.0194

Graph 9 represents the Score, RMSE, and MAE analysis of the Machine Learning Models used on
the dataset using SLR, DTR, RFR, KNNR, and SVR. These ML models are being implemented on the
dataset to train the model on the given dataset and predict the apparent temperature for the test datasets.
Graph 9 is the visual analysis and representation of the results of the models. This is concluded that the
model that produces the highest score is DTR and RFR with the lowest error rate for predicting the apparent
temperature using the test dataset.

Result Comparision
1.2
1
0.8
0.6
0.4
0.2
0
SLR DTR RFR KNNR SVR

Score RMSE MAE

Graph 9: Representation of the Score, RMSE, MAE investigation of the ML Models used on the dataset.
According to the discussion of the RMSE, and MAE values, given in table 1 and represented in figure
9. The best machine learning models for the data analysis in the proposed framework are random forest
regression and decision tree regression as both models are having highest score value (0.99) and lowest
RMSE (DTR: 0.0439, and RFR: 0.0427)and MAE (RMSE: 0.0019 and MAE: 0.0018) in comparison with
the other Model's simple linear regression; the score is 0.9901, RMSE is 0.9900 and MAE is 0.0099, K-
nearest neighbor score is 0.972, RMSE is 0.0285, MAE is 0.169 and support vector regression score is
0.9966, RMSE is 0.1394, MAE is 0.0194 obtained while implementing the ML models using the proposed
architecture. After the whole study, it is observed that the apparent temperature gets affected (increase or
decrease) because of the other parameters like temperature, humidity, pressure, wind, etc., and results in
increased energy consumption. Thus, the proposed architecture is helpful to have a watch on the different
parameters which cause high energy consumption. After that, this system will help manage the different
parameters accordingly and reduce energy consumption.

8. Comparison of the Proposed Work with Existing Work

In this section, the proposed work is being compared with the existing work. In 2020, Hoque et al.
proposed a novel regression-based ensemble forecast system to forecast the energy requirement. The author

146
addresses the enigma as regression modeling. The DS tool named RapidMiner is used for all data pre-
processing computations. The author used the dataset available publicly from Kaggle, which was having
31 features gathered from different sensors mounted in smart homes. Out of all the features the author
utilized 23 features for regression modeling. The model describes the 0.998 corrections between the features
and their label and achieved 38.15 RMSE value [52]. The author also implemented several other ML models
for regression such as SVM, RFR, etc., to apprehend their performance against the RMSE given in table 2.
The author clearly stated that the their proposed work requires enhancement, particularly for RMSE value.
The proposed HEMS is having much lower RMSE than the existing model. The RMSE values obtained in
the proposed and the existing work are represented in Table 2 and Graph 10 represents the improvement in
the proposed work and the existing work.
Table 2:
The RMSE values of the proposed and the Existing Work
Machine Learning Algorithm RMSE(Proposed) RMSE (Existing)
Simple Linear Regression (SLR) 0.9900 38.15
Support Vector Machine Regression (SVR) 0.1394 113.85
Random Forest Regression 0.0427 9.165

Comparision of Proposed and the

Existing Work
150

100

50

0
SLR SVM RFR

Proposed Existing

Graph 10: Representation of the comparison of the RMSE values of Proposed and Existing Work
A significant improvement is visible in Graph 10 in comparison of the existing work. The proposed work
is having less RMSE which means the error in the prediction of the values is reduced significantly. Thus,
the proposed work is improving the existing work as represented in section 7.

9. Conclusion with Future Work

At present, the energy requirement is growing day by day. With the increasing demand to meet the
requirements of the consumers, the most prior concern of our administration/government is to look into the
matter. This is possible to tackle this challenge with the help of modern technology. In a smart home,
temperature, humidity, wind speed, air pressure, etc. are some primary parameters to predict the apparent
temperature. The data of these parameters are being given to the machine learning models to predict the
apparent temperature. In this paper, the regression models (DTR, RFR, SLR, KNNR, SVR) are applied to
the dataset to train the ML model and to make the prediction of the apparent temperature, we obtain that
the DTR and RFR are giving the highest score and lowest error rates represented in figure 9. This shows
that these models are predicting the apparent temperature more accurately than the other models i.e.,
KNNR, SVR, and SLR. To conclude that analyzing and implementing the most affecting parameters which

147
affect the increase or decrease of the demand of energy and consumption of energy in the proposed
architecture can be done with the help of DTR and RFR. The existing architectures do not suggest the
patterns which can be used for the recommendation system in HEMS. But in the proposed architecture,
these systems are being introduced, which increase the system's impact and help to decrease the energy
usages in the home environment. The future direction of the proposed work is, to use the real-time data of
smart homes and optimized patterns of energy consumption in the obtained data using the machine learning
algorithms on the administration layer. The obtained patterns will be provided to a personalized
recommendation system on the service and security layer. This system will provide a recommendation to
the users and will assistance in decreasing the energy usages as well as the bills of the residents in the smart
homes. The aim of the system is to reduce the energy usages and electricity bills without interrupting the
comfort zones of the residents.

10. References
[1] Sampathkumar, A., Murugan, S., Sivaram, M., Sharma, V., Venkatachalam, K., & Kalimuthu, M. (2020).
Advanced Energy Management System for Smart City Application Using the IoT. Internet of Things in Smart
Technologies for Sustainable Urban Development, 185-194.
[2] Cho, K., Jang, H., Park, L. W., Kim, S., & Park, S. (2019, January). Energy management system based on
augmented reality for human-computer interaction in a Smart City. In 2019 IEEE International Conference on
Consumer Electronics (ICCE) (pp. 1-3). IEEE.
[3] Kim, D., Kwon, D., Park, L., Kim, J., & Cho, S. (2020). Multiscale LSTM-Based Deep Learning for Very-Short-
Term Photovoltaic Power Generation Forecasting in Smart City Energy Management. IEEE Systems Journal, 15(1),
346-354.
[4] Canizes, B., Pinto, T., Soares, J., Vale, Z., Chamoso, P., & Santos, D. (2017, June). Smart city: A GECAD-BISITE
energy management case study. In International Conference on Practical Applications of Agents and Multi-Agent
Systems (pp. 92-100). Springer, Cham.
[5] Canizes, B., Pinto, T., Soares, J., Vale, Z., Chamoso, P., & Santos, D. (2017, June). Smart city: A GECAD-BISITE
energy management case study. In International Conference on Practical Applications of Agents and Multi-Agent
Systems (pp. 92-100). Springer, Cham.
[6] Golpîra, H., & Bahramara, S. (2020). Internet-of-things-based optimal smart city energy management considering
shiftable loads and energy storage. Journal of Cleaner Production, 264, 121620.
[7] Mariano-Hernández, D., Hernández-Callejo, L., Zorita-Lamadrid, A., Duque-Pérez, O., & García, F. S. (2021). A
review of strategies for building energy management system: Model predictive control, demand side management,
optimization, and fault detect & diagnosis. Journal of Building Engineering, 33, 101692.
[8] Jasim, N. A., TH, H., & Rikabi, S. A. (2021). Design and Implementation of Smart City Applications Based on
the Internet of Things. International Journal of Interactive Mobile Technologies, 15(13).
[9] Wang, D., Zhong, D., & Souri, A. (2021). Energy management solutions in the Internet of Things applications:
Technical analysis and new research directions. Cognitive Systems Research, 67, 33-49.
[10] Lim, C., Cho, G. H., & Kim, J. (2021). Understanding the linkages of smart-city technologies and applications:
Key lessons from a text mining approach and a call for future research. Technological Forecasting and Social
Change, 170, 120893.
[11] Razmjoo, A., Nezhad, M. M., Kaigutha, L. G., Marzband, M., Mirjalili, S., Pazhoohesh, M., ... & Piras, G. (2021).
Investigating smart city development based on green buildings, electrical vehicles and feasible
indicators. Sustainability, 13(14), 7808.
[12] Al Sharif, R., & Pokharel, S. (2021). Smart City Dimensions and Associated Risks: Review of
literature. Sustainable Cities and Society, 103542.
[13] Li, J., Dai, J., Issakhov, A., Almojil, S. F., & Souri, A. (2021). Towards decision support systems for energy
management in the smart industry and Internet of Things. Computers & Industrial Engineering, 161, 107671.

148
[14] Francisco, A., Mohammadi, N., & Taylor, J. E. (2020). Smart city digital twin–enabled energy management:
Toward real-time urban building energy benchmarking. Journal of Management in Engineering, 36(2), 04019045.
[15] Nikitas, A., Michalakopoulou, K., Njoya, E. T., & Karampatzakis, D. (2020). Artificial intelligence, transport
and the smart city: Definitions and dimensions of a new mobility era. Sustainability, 12(7), 2789.
[16] Sivapragash, C., Padmanaban, S., Eklas, H., Holm-Nielsen, J. B., & Hemalatha, R. (2019). Location-based
optimized service selection for data management with cloud computing in smart grids. Energies, 12(23).
https://doi.org/10.3390/en12234517

[17] Yu, L., Xie, W., Xie, D., Zou, Y., Zhang, D., Sun, Z., Zhang, L., Zhang, Y., & Jiang, T. (2020). Deep
Reinforcement Learning for Smart Home Energy Management. IEEE Internet of Things Journal, 7(4), 2751–2762.
https://doi.org/10.1109/JIOT.2019.2957289

[18] Zafar, U., Bayhan, S., & Sanfilippo, A. (2020). Home Energy Management System Concepts, Configurations,
and Technologies for the Smart Grid. IEEE Access, 1–1. https://doi.org/10.1109/access.2020.3005244

[19] Dinh, H. T., Yun, J., Kim, D. M., Lee, K. H., & Kim, D. (2020). A Home Energy Management System with
Renewable Energy and Energy Storage Utilizing Main Grid and Electricity Selling. IEEE Access, 8, 49436–49450.
https://doi.org/10.1109/ACCESS.2020.2979189

[20] Al-Ali, A. R., Zualkernan, I. A., Rashid, M., Gupta, R., & Alikarar, M. (2017). A smart home energy
management system using IoT and big data analytics approach. IEEE Transactions on Consumer Electronics, 63(4),
426–434. https://doi.org/10.1109/TCE.2017.015014

[21] Mohapatra, B. N., & Panda, P. P. (2019). Machine learning applications to smart city. ACCENTS Transactions
on Image Processing and Computer Vision, 5(14), 1.

[22] Maulud, D., & Abdulazeez, A. M. (2020). A Review on Linear Regression Comprehensive in Machine
Learning. Journal of Applied Science and Technology Trends, 1(4), 140-147.

[23] Kumar, H., Singh, M. K., Gupta, M. P., & Madaan, J. (2020). Moving towards smart cities: Solutions that lead
to the Smart City Transformation Framework. Technological forecasting and social change, 153, 119281.

[24] Sampathkumar, A., Maheswar, R., Harshavardhanan, P., Murugan, S., Jayarajan, P., & Sivasankaran, V. (2020,
July). Majority voting-based hybrid ensemble classification approach for predicting parking availability in smart city
based on IoT. In 2020 11th International Conference on Computing, Communication and Networking Technologies
(ICCCNT) (pp. 1-8). IEEE.

[25] Link of the Dataset used in this paper: https://www.kaggle.com/budincsevity/szeged-weather

[26] Shafiq, M., Tian, Z., Sun, Y., Du, X., & Guizani, M. (2020). Selection of effective machine learning algorithm
and Bot-IoT attacks traffic identification for internet of things in smart city. Future Generation Computer
Systems, 107, 433-442.

[27] Shahabi, H., Shirzadi, A., Ghaderi, K., Omidvar, E., Al-Ansari, N., Clague, J. J., ... & Ahmad, A. (2020). Flood
detection and susceptibility mapping using sentinel-1 remote sensing data and a machine learning approach: Hybrid
intelligence of bagging ensemble based on k-nearest neighbor classifier. Remote Sensing, 12(2), 266.

[28] Luckey, D., Fritz, H., Legatiuk, D., Dragos, K., & Smarsly, K. (2020, August). Artificial intelligence techniques
for smart city applications. In International Conference on Computing in Civil and Building Engineering (pp. 3-15).
Springer, Cham.

149
[29] Ahmad, W., Ayub, N., Ali, T., Irfan, M., Awais, M., Shiraz, M., & Glowacz, A. (2020). Towards short term
electricity load forecasting using improved support vector machine and extreme learning machine. Energies, 13(11),
2907.

[30] Sivaram, M., Lydia, E. L., Pustokhina, I. V., Pustokhin, D. A., Elhoseny, M., Joshi, G. P., & Shankar, K. (2020).
An optimal least square support vector machine-based earnings prediction of blockchain financial products. IEEE
Access, 8, 120321-120330.

[31] Tong, Z., Miao, J., Tong, S., & Lu, Y. (2021). Early prediction of remaining useful life for Lithium-ion batteries
based on a hybrid machine learning method. Journal of Cleaner Production, 317, 128265.

[32] Ahmad, F., Almuayqil, S. N., Mamoona, H., Shahid, N., Wasim Ahmad, K., & Kashaf, J. (2021). Prediction of
COVID-19 cases using machine learning for effective public health management. Computers, Materials, & Continua,
2265-2282.

[33] Ahmad, A., Ostrowski, K. A., Maślak, M., Farooq, F., Mehmood, I., & Nafees, A. (2021). Comparative study of
supervised machine learning algorithms for predicting the compressive strength of concrete at high
temperature. Materials, 14(15), 4222.

[34] Angelaki, A., Singh Nain, S., Singh, V., & Sihag, P. (2021). Estimation of models for cumulative infiltration of
soil using machine learning methods. ISH Journal of Hydraulic Engineering, 27(2), 162-169.

[35] Sahoo, A., Samantaray, S., & Ghose, D. K. (2021). Prediction of flood in Barak River using hybrid machine
learning approaches: a case study. Journal of the Geological Society of India, 97(2), 186-198.

[36] Watt‐Meyer, O., Brenowitz, N. D., Clark, S. K., Henn, B., Kwa, A., McGibbon, J., ... & Bretherton, C. S. (2021).
Correcting weather and climate models by machine learning nudged historical simulations. Geophysical Research
Letters, 48(15), e2021GL092555.

[37] Tarana Singh, Arun Solanki, Anand Nayyar, “Multilingual Opinion Mining Recommendation System Using
RNN for Movie Domain” Accepted for publication in International conference in Computing, Communication and
Cyber-security (IC4S) 12-13 October, 2019, Chandigarh.

[38] R. Rajput and A. Solanki, “Real Time Analysis of Tweets using Machine Learning and Semantic Analysis”,
International Conference on Communication and Computing Systems(ICCCS-2016), Taylor and Francis, at
Dronacharya College of Engineering, Gurgaon, 9-11 September, pp. 687-692, 2016.

[39] Akash Tayal, Arun Solanki and Simarpreet Singh, ” Integrated Frame work for Identifying Sustainable
Manufacturing Layouts based on Big Data, Machine Learning, Meta-Heuristic and Data Envelopment Analysis “,
Accepted for Publication in Journal Sustainable Cities and Society, Elsevier Publication, IF=5.2 SCIE.

[40] Solanki, A., & Nayyar, A. (2019). Green internet of things (G-IoT): ICT technologies, principles, applications,
projects, and challenges. In Handbook of research on big data and the IoT (pp. 379-405). IGI Global.

[41] Nayyar, A., Rameshwar, R. U. D. R. A., & Solanki, A. (2020). Internet of Things (IoT) and the digital business
environment: a standpoint inclusive cyber space, cyber crimes, and cybersecurity. The Evolution of Business in the
Cyber Age, 10, 9780429276484-6.

[42] Singh, T., Solanki, A., & Sharma, S. K. (2021). Role of Smart Buildings in Smart City—Components,
Technology, Indicators, Challenges, Future Research Opportunities. Digital Cities Roadmap: IoT‐Based Architecture
and Sustainable Buildings, 449-476.

150
[42] Singh, S. P., Solanki, A., Singh, T., & Tayal, A. (2021). Internet of Intelligent Things: Injection of Intelligence
into IoT Devices. In Artificial Intelligence to Solve Pervasive Internet of Things Issues (pp. 85-102). Academic Press.

[43] M. U. Saleem, M. R. Usman and M. Shakir, "Design, Implementation, and Deployment of an IoT Based Smart
Energy Management System," in IEEE Access, vol. 9, pp. 59649-59664, 2021, doi: 10.1109/ACCESS.2021.3070960.

[44] Chouaib, B., Lakhdar, D., & Lokmane, Z. (2019, March). Smart home energy management system architecture
using IoT. In Proceedings of the 9th International Conference on Information Systems and Technologies (pp. 1-5).

[45] Al-Ali, A. R., Zualkernan, I. A., Rashid, M., Gupta, R., & AliKarar, M. (2017). A smart home energy management
system using IoT and big data analytics approach. IEEE Transactions on Consumer Electronics, 63(4), 426-434.

[46] Machorro-Cano, I., Alor-Hernández, G., Paredes-Valverde, M. A., Rodríguez-Mazahua, L., Sánchez-Cervantes,
J. L., & Olmedo-Aguirre, J. O. (2020). HEMS-IoT: A big data and machine learning-based smart home system for
energy saving. Energies, 13(5), 1097.

[47] Li, W., Logenthiran, T., Phan, V. T., & Woo, W. L. (2018). Implemented IoT-based self-learning home
management system (SHMS) for Singapore. IEEE Internet of Things Journal, 5(3), 2212-2219.

[48] Zafar, U., Bayhan, S., & Sanfilippo, A. (2020). Home energy management system concepts, configurations, and
technologies for the smart grid. IEEE access, 8, 119271-119286.

[49] Dinh, H. T., Yun, J., Kim, D. M., Lee, K. H., & Kim, D. (2020). A home energy management system with
renewable energy and energy storage utilizing main grid and electricity selling. IEEE Access, 8, 49436-49450.

[50] Zekić-Sušac, M., Mitrović, S., & Has, A. (2021). Machine learning based system for managing energy efficiency
of public sector as an approach towards smart cities. International journal of information management, 58, 102074.

[51] Hoque, M. S., Jamil, N., Saharudin, S. A., & Amin, N. (2020, August). Energy Usage Prediction for Smart Home
with Regression Based Ensemble Model. In 2020 8th International Conference on Information Technology and
Multimedia (ICIMU) (pp. 378-383). IEEE.

151