Using Data Mining to Identify COSMIC Function Point Measurement Competence

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 8, No. 6, December 2018, pp. 5253~5259
ISSN: 2088-8708, DOI: 10.11591/ijece.v8i6.pp5253-5259  5253
Journal homepage: http://iaescore.com/journals/index.php/IJECE
Using Data Mining to Identify COSMIC Function Point
Measurement Competence
Selami Bagriyanik1
, Adem Karahoca2
1
Digital Learning Solutions Technology Department, Turkcell Technology, Turkey
2
Department of Software Engineering, Bahcesehir University, Turkey
Article Info ABSTRACT
Article history:
Received Feb 27, 2018
Revised Jun 18 2018
Accepted Jul 29, 2018
Cosmic Function Point (CFP) measurement errors leads budget, schedule and
quality problems in software projects. Therefore, it’s important to identify
and plan requirements engineers’ CFP training need quickly and correctly.
The purpose of this paper is to identify software requirements engineers’
COSMIC Function Point measurement competence development need by
using machine learning algorithms and requirements artifacts created by
engineers. Used artifacts have been provided by a large service and
technology company ecosystem in Telco. First, feature set has been extracted
from the requirements model at hand. To do the data preparation for
educational data mining, requirements and COSMIC Function Point (CFP)
audit documents have been converted into CFP data set based on the
designed feature set. This data set has been used to train and test the machine
learning models by designing two different experiment settings to reach
statistically significant results. Ten different machine learning algorithms
have been used. Finally, algorithm performances have been compared with a
baseline and each other to find the best performing models on this data set. In
conclusion, REPTree, OneR, and Support Vector Machines (SVM) with
Sequential Minimal Optimization (SMO) algorithms achieved top
performance in forecasting requirements engineers’ CFP training need.
Keyword:
COSMIC funtion point
Educational data mining
Machine learning algorithms
Requirements analyst
Requirements artifacts
Copyright © 2018 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Adem Karahoca,
Department of Software Engineering,
Bahcesehir University,
Faculty of Engineering, Besiktas, Istanbul, 34349, Turkey.
Email: adem.karahoca@eng.bau.edu.tr; akarahoca@gmail.com
1. INTRODUCTION
Requirements engineers are one of the key profiles within software development teams. They
balances all project stakeholders’ expectations from idea to post production phases. Requiremens engineering
is not solely a technical discipline. Additionally, it also has an inter-disciplinary nature that concerns
Cognitive Psychology, Anthropology, Sociology, Linguistics and Philosophy aspects of the subject [1]. Thus,
they have a dramatical impact on the success of the software products and their continuous competence
development is critical.
Functional size measurement (FSM) is an important task that is used for scoping, budgeting,
managing outsourcing contracts, effort estimation, etc. This task is generally under the responsibility of
system analysts or requirements engineers (REs). CFP is one of the recent FSM methods. Function Point
variants are mainly used in software cost estimation [2] and productivity of the development organisations
[3]. Function Point is also a good indicator in identifying business complexity of the software [4].
Additionally It’s CFP variant is also a strong tool for requirement quality and process improvement [5]. CFP
measurement errors, made by requirements engineers, leads budget, schedule and quality problems in
software projects.

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 8, No. 6, December 2018: 5253 - 5259
5254
Therefore, it’s crucial to foresee and plan requirements engineers’ CFP training need in a quick and
correct manner. A recent paper points out that CFP training need should be represented more in higher
education [6]. We think training is also critical in the workplace setting and REs should be continually
developed in CFP competence when need arises. Training is the dominating factor for quality improvement
of FSM [7]. Factors that cause inconsistent and inaccurate CFP measurements might be improved by training
[8]. In this study, requirements engineers CFP training need has been forecasted by using the artifacts they
produced in the workplace and machine learning algorithms.
Data mining ors software analytics studies that use requirements engineering artifacts are scarce
[9]. For example, one of these rare studies aims early test effort prediction by using UML diagrams [10]. On
the other hand, software data mining studies which are based on sofware code and code change artifacts are
common [9]. For instance, code smells in the source code have been investigated using Neural Network
Models in a recent study [11]. We observe that using CFP data and data mining for Educational purposes is
even more rare in the literature. As far as we know, this research is the first study in the literature that uses
CFP data and educational data mining to improve REs’ CFP measurement capabilities. The rest of the paper
is organized as follows: in the 2nd section, a background on Data Mining, Machine Learning Algorithms,
CFP and study details are provided. Results are presented in the 3rd section and is followed by conclusions in
the 4th section.
2. RESEARCH METHOD
In this section, first of all, machine learning and CFP methods are explained briefly in the
subsections 2.1 and 2.2. Second, CFP training need prediction usecase, feature set design and data gathering
and preparation phases of the study is presented in 2.3, 2.4 and 2.5. Finally, models training details and
evaluation results are given in 2.6.
2.1. Data Mining and Machine Learning Algorithms
Data Mining is defined as “the process of discovering patterns, automatically or semi-automatically,
in large quantities of data” [12]. Knowledge discovery from data (KDD) is another common term used in the
literature [13]. Following algorithms which were implemented in Weka [14] are used in this study:
 Random Forest (RF): This is an ensemble learning method consisting of a set of decision tree
classifiers. Each tree in the forest is triggered by an independently created random number vector [15].
 Naïve Bayes (NB): This method uses Bayes’ rule to do the classification by computing class
probabilities and using observed attribute values. The method is called “naïve” since it has two basic
assumptions: attributes are conditionally independent and no hidden factor impacts on the prediction
process [16].
 REPTree: This is a fast decision tree algorithm that generates a decision tree using information gain
method to split [17]. Missing values are managed as in C4.5 algorithm [18].
 J48: It is a Java implementation of a slightly different version of C4.5 [17].
 LMT: Logistic Model Trees are standard decision trees which use logistic regression functions at their
leaves [19].
 Multilayer Perceptron (MLP): MLP is a feed-forward artificial neural network which uses back
propagation training algorithm. It is a system of interconnected nodes or neurons which maps an input
vector into an output vector to maintain a nonlinear relation [20]. The neurons are connected via
weights and output signals [20].
 Support Vector Machines (SVM) and Sequential Minimal Optimization (SMO): In linear case, an
SVM is a hyperplane that set a boundary between some positive instances and negative instances [21].
It can also be further extended to non-linear cases [21]. Training an SVM requires quadratic
programming (QP) optimization problem solving which is a very time and memory consuming
operation and SMO is a substantial improvement on the original training algorithm [21].
 K-nearest Neighbour Classifier (IBk): It classifies a data point based on its k most similar other data
points [22].
 ZeroR: It predicts the majority class of nominal test data while it predicts the average value if numeric
class is the case [12]. In this study, it will be used as a baseline for the performance of machine learning
algorithms.
 OneR: This method classifies instances based on a one rule which is extracted from a single
attribute [23].

Int J Elec & Comp Eng ISSN: 2088-8708 
Using Data Mining to Identify COSMIC Function Point Measurement Competence (Selami Bagriyanik)
5255
2.2. COSMIC Function Point (CFP)
Software functional size measurement (FSM) has been in use for more than forty years [24]. There
are many FSM methods [25]. COSMIC Function Point is a new generation software functional size
measurement method. First version of the method was published in 1988 [26]. It’s one of four ISO certified
FSM methods which are dominating in the industry: IFPUG, COSMIC, NESMA and Mark II [25]. CFP
measurement is a three-step process: measurement strategy, mapping and measurement steps. Purpose, scope,
and level of granularity of the measurement are determined in the first step; in mapping phase, functional
processes and data groups in the requirements are determined; in the final stage, data movements are
specified and counted, for all functional processes [26]. CFP is the functional sizing measurement method
that is used in the company under study [3,7,27].
2.3. CFP Requirement Ontology
Requirement artefacts that will be used in training the machine learning models are instances of
requirement and CFP ontologies designed in [3]. Currently, this requirement ontology and CFP
measurements are standard methods used by requirements engineers within the same telecommunications
company in which this study is conducted. Periodically, a subset of all requirements documents are randomly
selected and examined by internal audit team manually to identify errors in CFP measurements. After each
audit, problematic CFP measurements are identified, recorded and potential learning needs are reported to
requirements engineering management. By this data mining research, the manual examination process by
audit team is intended to be semi-automated and learning opportunities will automatically be extracted from
requirements documents.
2.4. Feature Set Extraction from CFP Requirement Ontology
CFP Ontology concepts are shown in the second column of the Table 1. Related concepts are
categorised into concept categories to specify data indicators that will be used in data mining process.
Ontology concept categories are shown in the first column of Table 1. As a result, features of the data and the
predicted outcome (Class) of the classification process is shown Table 2. In Table 2, the first seven attributes
are input attributes and the last one, “CFP Training Need”, is class or Classification result.
Table 1. CFP Ontology Concept Categories
Ontology Concept Category Ontology Concept
Use case Use case
Use case Application Interaction Diagram
Interaction Interaction
Evolution Type Add Evolution Type
Evolution Type Modify Evolution Type
Evolution Type Delete Evolution Type
Application Application Business Module
Application Application Database Module
Application Application Service
Application Application Service Boundary
Use case Use case Actor
Use case Use case Event
Information Information Asset
Interaction Integration Entry Interaction
Interaction Integration Exit Interaction
Interaction User Interface Entry Interaction
Interaction User Interface Exit Interaction
Interaction Database Write Interaction
Interaction Database Read Interaction
Scope Project Scope
Scope Application Service Scope
Not Applicable Productivity Measurement
Business Logic Use case Business Logic
Business Logic Interaction Business Logic
Table 2. CFP Ontology Indicator set and Their Possible Values
Ontology Concept Category Value
Use case Yes, No, Partial
Interaction Yes, No, Partial
Information Yes, No, Partial
Evolution Type Yes, No, Partial
Business Logic Yes, No, Partial
Application Yes, No, Partial
Scope Yes, No, Partial
CFP Training Need Yes, No

 ISSN: 2088-8708
5256
2.5. Data Gathering and Preparation
First seven data attributes shown in Table 2 are obtained from requirements documents by checking
whether each concept category value is existing or not. If their values are partially existing then the concept
category value is recorded as “Partial”. The last attribute is captured from audit examination results. If the
difference between the actual measurement done by requirements engineer and correct measurement result
identified by audit team is greater than 5 % then this case is recorded as a learning opportunity by recording
“CFP Training Need” value as “Yes”. 101 data points have been collected and results have been recorded in
a comma-delimited values (.csv) file. Next, this file has been converted into the attribute-relation file format
(.arff) which is the standard file format used by The Waikato Environment for Knowledge Analysis (WEKA)
data mining software [14]. The conversion tool used for .csv to .arff is an online web tool [28].
2.6. Model Training and Evaluation
To train and evaluate the machine learning models, Weka Experimenter [29] has been used. Two
experiments have been done in Weka. In the first experiment, “Data sets first” parameter checked and
number of repetitions has been set as 100 in iteration control parameters panel. Experiment type is selected as
Cross validation. Number of folds attribute is set to 10. Dataset has been selected as the .arff file which is
created as described in section 2.5. All algorithms which have been explained in section 2.1have been
selected in Algorithms panel. Next, the experiment with this configuration has been run on an Intel Core i7-
5600U CPU, 2.6 GHz, 8 GB RAM and 64-bit Windows Operating System machine.
The total execution time was 194 seconds and MLP had the slowest running time. Finally, in
analyse tab of Weka Experimenter user interface, all algorithms have been selected as test base separately
and test is performed for each algorithm. Test has been repeated for three evaluation metrics: Accuracy
(Number of Correct Classifications), F-Measure (FM), and Kappa statistic. In the second experiment,
experiment type was set to Train/Test Percentage Split (data randomized) and train percentage was set to
66%. All other configuration remained the same as in Experiment 1. In this case, the total execution time was
74 seconds and MLP had the slowest running time, again.
3. RESULTS AND ANALYSIS
Table 3 designates the algorithm performances in terms of accuracy, FM, and Kappa metrics for
both experiments. Metric values are shown as averages with standard deviations. All algorithms seem
meaningfully better than ZeroR baseline performance. Support Vector Machines and OneR algorithms have
the largest average accuracy values. However evaluating the algorithms solely based on the average values
and standard deviations wouldn’t be sufficient since differences between results might not be statistically
significant. Therefore, in Weka Experimenter Analyse interface, Significance has been set to 0.05, all
algorithms have been selected as test base separately and tests have been performed. Statistically significant
differences have been recorded during tests. Statistical significance is denoted by “v” and “*” symbols in the
Weka interface. Former means statistically significant better performance while latter implies statistically
significant worse performance [29].
Statistically significant superiorities between algorithms are shown in Table 4. For instance, in
Experiment 1, Naïve Bayes performs better than IBk and ZeroR when Accuracy and Kappa metrics are
concerned. Best performing algorithms have been determined by comparing the number of all statistically
significant superiorities. We show this number as “Number of Wins” in Table 4. As a result; REPTree, OneR
and SVM with SMO algorithms have the maximum “Number of Wins” values and are determined to be the
top three algorithms performing best in CFP dataset of this study. As far as we know, this is the first study
that use data mining on requirements and CFP measurement data. Therefore, we couldn’t compare the
performance of our study with other similar research directly. However if we benchmark with some
educational data mining studies in general, we see our top performing models are very good in terms of
accuracy [30–33].
Table 3. Algorithm Performances for CFP Dataset
Algorithm
Experiment 1: Cross - validation Experiment 2: 66% Split Test
Accuracy (%) FM Kappa Accuracy FM Kappa
Random Forest (RF) 78.79 (11.72) 0.82 (0.11) 0.56 (0.25) 79.20 (5.54) 0.83 (0.04) 0.56 (0.12)
Naive Bayes (NB) 82.97 (11.13) 0.86 (0.09) 0.63 (0.24) 81.29 (5.46) 0.85 (0.04) 0.60 (0.12)
REPTree 84.02 (11.25) 0.86 (0.11) 0.67 (0.23) 84.27 (4.81) 0.86 (0.04) 0.68 (0.10)
J48 (Weka C 4.5 Implementation) 82.59 (11.39) 0.84 (0.11) 0.65 (0.23) 83.38 (4.58) 0.85 (0.04) 0.66 (0.09)
Logistic Model Trees (LMT) 83.84 (11.34) 0.86 (0.11) 0.67 (0.23) 83.52 (5.21) 0.86 (0.05) 0.66 (0.11)
Multilayer Perceptron (MLP) 76.74 (12.50) 0.80 (0.12) 0.52 (0.26) 75.94 (6.55) 0.80 (0.05) 0.49 (0.15)

5257
Table 3. Algorithm Performances for CFP Dataset
Algorithm
Experiment 1: Cross - validation Experiment 2: 66% Split Test
Accuracy (%) FM Kappa Accuracy FM Kappa
Support Vector Machines with SMO 84.16 (11.25) 0.86 (0.11) 0.68 (0.23) 83.70 (4.91) 0.86 (0.04) 0.67 (0.10)
K-nearest Neighbour Classifier (IBK) 75.61 (11.56) 0.80 (0.10) 0.48 (0.25) 75.38 (5.16) 0.81 (0.04) 0.47 (0.12)
OneR 84.16 (11.25) 0.86 (0.11) 0.68 (0.23) 84.27 (4.81) 0.86 (0.04) 0.68 (0.10)
ZeroR 59.45 (01.64) 0.75 (0.01) 0.00 (0.00) 59.37 (0.63) 0.75 (0.04) 0.00 (0.00)
Table 4. Statistically Significant Superiorities of Algorithms for CFP Dataset
Algorithm
Experiment 1: Cross - validation Experiment 2: 66% Split Number
of WinsAccuracy FM Kappa Accuracy FM Kappa
Random Forest (RF) ZeroR ZeroR ZeroR ZeroR ZeroR ZeroR 6
Naive Bayes (NB) IBk, ZeroR
MLP, IBk,
ZeroR
IBk,
ZeroR
ZeroR ZeroR ZeroR 10
REPTree
RF, MLP,
IBk, ZeroR
MLP, IBk,
ZeroR
RF,
MLP,
IBk,
ZeroR
IBk,
ZeroR
IBk,
ZeroR
IBk,
ZeroR
17
J48 (Weka C 4.5
Implementation)
IBk, ZeroR ZeroR
IBk,
ZeroR
IBk,
ZeroR
ZeroR
IBk,
ZeroR
10
Logistic Model Trees (LMT)
MLP, IBk,
ZeroR
MLP, IBk,
ZeroR
MLP,
IBk,
ZeroR
IBk,
ZeroR
ZeroR
IBk,
ZeroR
14
Multilayer Perceptron (MLP) ZeroR ZeroR ZeroR ZeroR ZeroR ZeroR 6
Support Vector Machines with
SMO
RF, MLP,
IBk, ZeroR
MLP, IBk,
ZeroR
RF,
MLP,
IBk,
ZeroR
IBk,
ZeroR
ZeroR
IBk,
ZeroR
16
K-nearest Neighbour Classifier
(IBk)
ZeroR ZeroR ZeroR ZeroR ZeroR ZeroR 6
OneR
RF, MLP,
IBk, ZeroR
MLP, IBk,
ZeroR
RF,
MLP,
IBk,
ZeroR
IBk,
ZeroR
IBk,
ZeroR
IBk,
ZeroR
17
ZeroR None None None None None None 0
4. CONCLUSION
In this study, we conducted an educational data mining research. In the scope of this use case, a CFP
dataset which was collected from a large telecommunications services and technology company has been
analysed using 10 machine learning algorithms to identify CFP learning need of Requirements Engineers.
After two experiments, model performances are evaluated and top performer algorithms have been identified.
REPTree, OneR and SVM with SMO algorithms performed better than other algorithms in a statistically
significant manner. Top performing model prediction performances are sufficient to be used in the
production environment in the company. In the future, following research is planned:
 Dominating indicators in CFP measurement will be identified by using feature selection algorithms.
Some new indicators from the requirements artifacts may arise in this process.
 Data points number will be increased and the study will be replicated by also adding some other
algorithms such as Adaptive Neuro Fuzzy Inference System (ANFIS).
REFERENCES
[1] Nuseibeh B, Easterbrook S. “Requirements engineering: a roadmap”. In: Proceedings of the conference on The
future of Software engineering - ICSE ’00. 2000. p. 35–46.
[2] Naik P, Nayak S. “Insights on Research Techniques towards Cost Estimation in Software Design.” Int J Electr
Comput Eng (IJECE). 2017;7(5):2883–94.
[3] Bagriyanik S, Karahoca A. Automated COSMIC Function Point measurement using a requirements engineering
ontology. Inf Softw Technol. 2016;72:189–203.
[4] Fajar AN, Shofi IM. Reduced Software Complexity for E-Government Applications with ZEF Framework.
TELKOMNIKA (Telekommunication Computing, Electronics and Control). 2017;15(1):415–20.
[5] Trudel S, Turcotte A. “Combining Qualitative and Quantitative Software Process Evaluation : A Proposed
Approach”. Coll Econ Anal Ann. 2017;43:135–54.
[6] Symons C, Abran A, Ebert C, Vogelezang F. “Measurement of software size : advances made by the COSMIC
community”. In: 2016 Joint Conference of the International Workshop on Software Measurement and the
International Conference on Software Process and Product Measurement. IEEE; 2016. p. 75–86.
[7] Salmanoğlu M, Öztürk K, Bağrıyanık S, Ungan E, Demirörs O. “Benefits and challenges of measuring software
size: early results in a large organization”. In: 25th International Workshop on Software Measurement and 10th

 ISSN: 2088-8708
5258
International Conference on Software Process and Product Measurement. 2015.
[8] Ozkan B, Demirors O. “On the Seven Misconceptions about Functional Size Measurement”. In: 2016 Joint
Conference of the International Workshop on Software Measurement and the International Conference on Software
Process and Product Measurement. IEEE; 2016. p. 45–52.
[9] Bagriyanik S, Karahoca A. “Big data in software engineering: A systematic literature review”. Glob J Inf Technol.
2016;6(1):107–16.
[10] Sahoo P, Mohanty JR. “Early Test Effort Prediction using UML Diagrams’. Indones J Electr Eng Comput Sci
(IJEECS). 2017;5(1):220–8.
[11] Kim DK. “Finding Bad Code Smells with Neural Network Models”. Int J Electr Comput Eng. 2017;7(6):3613–21.
[12] H. Witten I, Frank E, A. Hall M. “Data mining: Practical machine learning tools and techniques”. 3rd ed. 2011.
[13] Han J, KamberMicheline. “Data Mining Concepts and Techniques”. 2nd ed. 2006.
[14] Hall M, Frank E, Holmes G, Pfahringer B, Reutemann P, Witten IH. “The WEKA data mining software: an
update”. ACM SIGKDD Explor Newsl. 2009;11(1):10–8.
[15] Breiman L. Random forests. Mach Learn. 2001;45(1):5–32.
[16] H. John G, Langley P. “Estimating continuous distributions in Bayesian classifiers”. In: Proceedings of the
Eleventh conference on Uncertainty in artificial intelligence. Morgan Kaufmann Publishers Inc.; 1995. p. 338–45.
[17] Zhao Y, Zhang Y. “Comparison of decision tree methods for finding active objects”. Adv Sp Res. 2008;
41(12):1955–9.
[18] Quinlan JR. C4. 5: programs for machine learning. 2014.
[19] Landwehr N, Hall M, Frank E. “Logistic model trees”. Mach Learn. 2005;59(1–2):161–205.
[20] Gardner M., Dorling S. Artificial neural networks (the multilayer perceptron)—a review of applications in the
atmospheric sciences. Atmos Environ [Internet]. 1998;32(14–15):2627–36. Available from:
http://www.sciencedirect.com/science/article/pii/S1352231097004470
[21] Platt JC. Sequential Minimal Optimization: A Fast Algorithm for Training Support Vector Machines. Adv kernel
methods [Internet]. 1998;185–208. Available from:
http://www.bradblock.com/Sequential_Minimal_Optimization_A_Fast_Algorithm_for_Training_Support_Vector_
Machine.pdf
[22] Aha DW, Kibler D, Albert MK. Instance-Based Learning Algorithms. Mach Learn. 1991;6(1):37–66.
[23] Holte RC. Very Simple Classification Rules Perform Well on Most Commonly Used Datasets. Mach Learn.
1993;11(1):63–90.
[24] Bundschuh M, Dekkers C. “The IT measurement compendium: estimating and benchmarking success with
functional size measurement”. Springer Science & Business Media; 2008.
[25] Jones C. “Applied Software Measurement Global Analysis of Productivity and Quality”. McGraw-Hill Education
Group; 2008.
[26] Consortium CSMI. “The COSMIC Functional Size Measurement Method Version 4.0.1 Measurement Manual
“[Internet]. [cited 2018 Jan 17]. Available from: https://cosmic-sizing.org/publications/measurement-manual-401/
[27] Bağrıyanık S, Karahoca A, Ersoy E.” Selection of a functional sizing methodology: A telecommunications
company case study”. Glob J Technol. 2015;7(7):98–108.
[28] Ilya.kuzovkin@gmail.com. Online converter from .csv to WEKA .arff [Internet]. [cited 2018 Jan 20]. Available
from: http://ikuz.eu/csv2arff/
[29] Scuse D, Reutemann P. “WEKA Experimenter Tutorial for Version 3-5-8”. 2008.
[30] Asif R, Merceron A, Ali SA, Haider NG. Analyzing undergraduate students â€TM
performance using educational
data mining. Comput Educ [Internet]. 2017;113:177–94. Available from:
http://dx.doi.org/10.1016/j.compedu.2017.05.007
[31] Kabakchieva D. “Predicting Student Performance by Using Data Mining Methods for Classification”. Cybern Inf
Technol. 2013;13(1):61–72.
[32] Ahmad F, Ismail NH, Aziz AA. “The Prediction of Students ’ Academic Performance Using Classification Data
Mining Techniques”. Appl Math Sci. 2015;9(129):6415–26.
[33] Kaur P, Singh M, Singh G. “Classification and prediction based data mining algorithms to predict slow learners in
education sector”. In: P3rd International Conference on Recent Trends in Computing 2015(ICRTC-2015)
[Internet]. Elsevier Masson SAS; 2015. p. 500–8. Available from: http://dx.doi.org/10.1016/j.procs.2015.07.372

5259
BIOGRAPHIES OF AUTHORS
Dr. Selami Bagriyanik holds a PhD in Software Engineering. He is interested in software
development, requirements engineering, software measurement, advanced learning technologies,
data mining and big data. He also works as the Digital Learning and Business Solutions
Technology Manager in Turkcell.
Dr. Adem Karahoca holds a PhD in Software Engineering. He is interested in human–computer
interaction, web based education systems, data mining, big data, and management information
systems. He has published articles at prestigious journals about use and data mining applications
of business information systems in health, tourism, and education.

Using Data Mining to Identify COSMIC Function Point Measurement Competence

More Related Content

What's hot (18)

Similar to Using Data Mining to Identify COSMIC Function Point Measurement Competence (20)

More from IJECEIAES (20)

Recently uploaded (20)

Using Data Mining to Identify COSMIC Function Point Measurement Competence