Scalable Computing: Practice and Experience, ISSN 1895-1767, http://www.scpe.

org
© 2024 SCPE. Volume 25, Issues 2, pp. 1276–1285, DOI 10.12694/scpe.v25i2.2213

SOFTWARE EFFORT ESTIMATION USING MACHINE LEARNING ALGORITHMS

KRUTI LAVINGIA ∗, RAJ PATEL †, VIVEK PATEL ‡, AND AMI LAVINGIA §

Abstract. Effort estimation is a crucial aspect of software development, as it helps project managers plan, control, and schedule
the development of software systems. This research study compares various machine learning techniques for estimating effort in
software development, focusing on the most widely used and recent methods. The paper begins by highlighting the significance
of effort estimation and its associated difficulties. It then presents a comprehensive overview of the different categories of effort
estimation techniques, including algorithmic, model-based, and expert-based methods. The study concludes by comparing methods
for a given software development project. Random Forest Regression algorithm performs well on the given dataset tested along
with various Regression algorithms, including Support Vector, Linear, and Decision Tree Regression. Additionally, the research
identifies areas for future investigation in software effort estimation, including the requirement for more accurate and reliable
methods and the need to address the inherent complexity and uncertainty in software development projects. This paper provides
a comprehensive examination of the current state-of-the-art in software effort estimation, serving as a resource for researchers in
the field of software engineering.

Key words: Software Engineering, Machine Learning, Effort Estimation

1. Introduction. A large volume of data is produced while software companies develop and generate
software [1]. From the requirements phase until maintenance, a unique collection of data is generated at each
stage of software development. In software development project management, key factors such as Lines of Code
(LOC), historical project data, team skill levels, team size, functional and non-functional requirements, project
phases, and development timelines are crucial for success. LOC quantifies code changes, while past project data
informs resource allocation. Team skills and size impact productivity, while requirements shape the project’s
direction [2]. Monitoring project phases and adhering to timelines ensures progress and identifies bottlenecks.
These elements collectively enable effective project management, guiding teams to deliver software solutions
that align with stakeholder expectations and meet project goals efficiently.
The data produced in software repositories is collected and maintained by software organizations as part
of their ongoing efforts to improve the software quality. Several data mining techniques are used to analyze the
vast amounts of data kept in software repositories to uncover new patterns and standout data points [3].
Two-thirds of all large projects greatly exceed their original projections and one-third of projects go over
budget and are delivered late, as claimed by surveys [4].
The technique of estimating the effort and resources needed to construct a software system is known as
software engineering cost estimation. Managers and stakeholders typically use this process to plan and budget
software development projects. Cost estimation methods can range from simple rule-of-thumb calculations to
more formal methods, such as parametric modelling or expert judgment. Factors that can influence the cost
of a software project include the size and complexity of the system, the development methodologies and tools
used, and the skill level of the development team [5]. It is essential to note that cost estimation is an iterative
process that needs to be repeatedly refined and updated as more information becomes available throughout the
project.
In software engineering, effort estimation is the process of predicting the number of human resources,
measured in person-hours or person-months, needed to complete any software development project. It is a
critical aspect of project management as it helps stakeholders to plan and budget for the project and to make

∗ Nirma University (kruti.lavingia@nirmauni.ac.in)

† Institute of technology, Nirma University (20bce218@nirmauni.ac.in)
‡ Institute of technology, Nirma University(20bce226@nirmauni.ac.in)
§ Sal College of Engineering (ami.lavingia@sal.edu.in)

1276
 Software Effort Estimation using Machine Learning Algorithms 1277

informed decisions about resource allocation and project scope [6].

Software engineering has a variety of effort estimating techniques, each having advantages and disadvantages.
Among the more popular techniques are:
1.1. Methods for effort estimation.
• Expert judgment: This method relies on the experience and expertise of individuals who have
previously worked on similar projects. It is often used as a starting point for effort estimation and can
provide a rough estimate of the required effort. However, it is subject to bias and can be affected by
an individual’s experience [7].
• Analogous estimation: This method uses the data from previous similar projects to estimate the
effort required for the current project. It is a quick and easy method, but it is not always accurate, as
the projects may not be entirely similar [8].
• Three-point estimation: This method uses the most likely, optimistic, and pessimistic estimates of
effort to generate a range of possible values. It helps generate a range of likely effort estimates, but it
is a relatively complex method [9].
• Parametric estimation: This method uses mathematical models to estimate the effort required for
a project. It is based on the project’s size, complexity, and other [10].
Estimating is crucial in project management, as inaccuracies in estimation can lead to poor project perfor-
mance, potentially resulting in project failure. One of the management factors that cause about 65% of lost
projects is poor estimation technique [11]. This study uses machine learning to create a model for software cost
estimation. As a result, this review aims to test if machine learning is a better technique than using traditional
methods to estimate software development effort or vice versa. Support machine learning algorithms such as
vector regression, regression algorithms such as simple linear regression, and decision tree-based regression are
applied in this study with the assistance of the Python programming language.
The remainder of this paper is broken up into related work that focuses on previous studies that have
been done in this particular area. The following section examines machine learning methods and software cost
estimation. They utilized machine-learning techniques, data sets, and evaluation standards are thoroughly
detailed in the next section. The comparison and in-depth analysis of the experimental results come before the
conclusion and section on future work.
However, we do wish to stress the purpose of this paper is to consider how different prediction systems
perform under the same conditions and how to evaluate them, not to argue in favor of any particular prediction
technique.
2. Related work. Numerous research has suggested various models for calculating the cost of the software.
To find alternatives, improve upon, or support existing models, multiple models have been proposed and
constructed [12]. Various new models have been developed and constructed to discover alternatives, improve
existing models, or assist current models. A well-known method for estimating software costs is the build cost
model.
2.1. Without Machine Learning. As introduced by Barry Boehm, the Constructive Cost Model (CO-
COMO) stands out as the predominant approach within the algorithmic methods category [13]. It relies on a
series of equations and parameters derived from past software project experiences for estimation, and its models
have garnered widespread practical acceptance. In the context of COCOMO, code size is measured in Thousand
Lines Of Code (KLOC), and effort is expressed in person-months. COCOMO is a valuable tool to gauge the
quality and effort required for software projects, as exemplified by its application in Manikavelan’s study [14],
providing approximate estimates within fixed time frames. Moreover, the authors in this particular research
extended COCOMO’s capabilities by incorporating the Gaussian Membership Function, revealing outstanding
performance of the fuzzy-COCOMO model in terms of reducing relative errors.
In the study conducted by Nandan and Deepak [15], a novel approach was employed. They utilized a
hybrid BATGSA algorithm to optimize the COCOMO model, drawing data from NASA databases. The
study comprehensively compared three distinct techniques implemented using MATLAB. The outcome was a
noteworthy decrease in normalized error with the updated COCOMO model.
Notably, the authors introduced an innovative hybrid strategy that amalgamated fuzzy clustering, ABE,
1278 Kruti Lavingia, Raj Patel, Vivek Patel, Ami Lavingia

and ANN approaches to enhance the accuracy of effort estimation. This novel approach entailed clustering all
projects within a newly established framework, effectively mitigating the influence of inconsistent and irrelevant
projects on projections. This research resulted in significant improvements, with an average enhancement of
0.25 in the first dataset and remarkable gains of 52 and 94 per cent in the second dataset, as demonstrated by
the prediction percentage (PRED) and mean magnitude of relative error (MMRE) performance indicators.

2.2. With Machine Learning. In their study, Shukla et al. leveraged the Desharnais dataset to explore
the performance of various machine learning models in estimating software project effort [16]. Notably, their
MLPNN model achieved an R2 value of 0.79380, surpassing other models like LR, SVM, and KNN. It effectively
explained 79% of the estimated variance, with only marginal differences (6-7%) in R2 values among them.
The research delved into the association between the most correlated elements by Pearson correlation and
the effort variable using seven machine learning methods, following an initial correlation analysis of each dataset
variable with the effort variable. Performance evaluation was based on error values [17].
Sarro introduced an effort prediction technique combining Confidence Interval Analysis and Mean Absolute
Error assessment [18]. This innovative approach demonstrated promise through trials involving over 700 soft-
ware programs, finding applications in diverse fields like pharmaceutical research, biochemistry, and computer
vision. The method selected feature subsets based on optimization techniques and transferred them to classifiers
(SVM, ANN, and Decision Tree) for classification and regression tasks involving two optimization algorithms
and three classifiers. This process, known as Feature Selection, yielded excellent results across various datasets.
In another study, 93 projects’ preprocessed COCOMO NASA benchmark data were employed to make
predictions using machine learning techniques like Naive Bayes, Logistic Regression, and Random Forests [19].
Performance evaluation metrics such as Classification Accuracy, Precision, Recall, and AUC were employed
following five-fold cross-validation. Each method outperformed the benchmark COCOMO model in production
prediction.
V. Anandhiin’s investigation focused on regression techniques, notably the M5 algorithm and Linear Re-
gression, for estimating software cost using the Constructive Cost Model dataset [20]. The results indicated
that the M5 method exhibited more minor errors, including the mean magnitude of the relative error and
Median magnitude of the relative error, compared to Linear Regression in prediction. These clear distinctions
highlight where different methods are introduced and provide insights into the authors’ approach and findings.

3. Machine Learning. In this section, methods and ML algorithms are discussed; after that information
about the dataset is stated along with its structure. Finally, the evaluation standards are the topic of conver-
sation. Below stated section provides a clear and concise summary of some machine learning algorithms that
are used to predict the effort for the project.
Machine learning techniques are increasingly thought to be crucial in research. The results of ML approaches
are consistently reliable, and they are frequently employed reliable in numerous studies. Using two machine
learning methods, Yeha and Deng provided a system to forecast the software product life cycle [21]. The study
provided a more accurate and adaptable model for estimating product costs.
To distinguish between different types of breast cancers, Aleriza and Bita, used support vector machines,
K-nearest neighbors, and neural network classifiers [22].
Other studies concentrated more on the environment for cost assessment and other relevant elements, such
as the software development life cycle relevant to the particular project. For instance, in 2018, research on the
impact of organizational factors were published by Rahikkala et al., which looked at how its many components
could potentially influence and improve the software cost planning process [23].

3.1. Linear Regression. Based on other attributes’ values, linear regression analysis will predict a vari-
able’s value [24]. Two types of variables are there in the algorithm, one dependent and another independent.
The dependent variable is predicted by an algorithm. The dependent variable is predicted using the inde-
pendent variable as a basis. Such analysis determines the coefficients of linear mathematical equations using
independent variables that may most effectively anticipate the value of the dependent variable. This algorithm
fits the output on a straight line to reduce the discrepancy between the actual and anticipated output. The
value of A (the dependent variable/attribute) is then estimated from B (the independent attribute/variable).
 Software Effort Estimation using Machine Learning Algorithms 1279

The technique used is straightforward and comparative. It is less complex than other methods for predic-
tions. The equation for linear regression:

Y = β0 + β1 X1 + β2 X2 + ... + βn Xn + e (3.1)

Y is the dependent variable and β1 ,β2 ,...,βn are coefficients and X1 , X1 , ... ,Xn are the independent
variables. Here β0 is the intercept of the line which is generated by the vertical axis. e is an error term; it is the
random error used to express some random factors’ effect of some random factors on Y (dependent variables).

3.2. Support Vector Regression. Drucker et al. initially presented the Support Vector Regression
(SVR) model for regression analysis in 1997 [25].
Convex quadratic programming issues are replaced with convex linear system solutions in the least-squares
SVR (LS-SVR) [26] [27], which greatly speeds up training. An extensive empirical investigation has revealed
that the LS-generalization SVR’s performance is comparable to the SVR’s. (Van Gestel et al., 2004) [26].
Among these, Vapnik’s SVR, which employs regression analysis via the support vector machine (SVM), has
a wide range of applications in the energy-prediction industries. The principle of structural risk minimization
underpins SVR. It has clear advantages for small datasets and can maintain excellent generalization ability [28].
In support vector regression, there are two types of hyperparameters, first being the Kernel function and
second is a variable C which is defined as the penalty parameter of the error term. Kernel parameters affect
the separation boundary. Since SVR is a kernel-based method, the performance of SVR heavily depends on
the kernel functions. Different kernel equations are there, which can be applied to get the different decision
boundaries.
These kernels project the input data into several high-dimensional feature spaces. Because high-dimensional
feature spaces perform so well, creating new kernel functions can surpass SVR performance thresholds.
The distribution and shape of the dataset affect how the kernel is used. Here, the objective function for
SVR’s normal vector’s size is |w|, which is minimized [26].

xi w + b = 0 (3.2)

xi w + b ≥ +1,yi = +1 (3.3)
xi w + b ≤ −1,yi = −1 (3.4)
yi (xi w + b) − 1 ≥ 0, ∀i (3.5)

∥w∥2
minimize 2 (3.6)
l l
X 1 X
Maximize W (α) = αi − αi αj yi yj (xi xj ) (3.7)
i=1
2 i,j=1

3.3. Random Forest Regression. Tin Kam Ho first introduced random forests in 1995. For the creation
of a decision tree, the Random forests method is applied, which solves problems of classification and regression.
Later, Breiman extended the technique by combining bagging with features selected at random [29], allowing
for the controlled construction of multiple decision trees using variance. Compared to decision trees, the random
forest algorithm provides more accurate error rate estimates. More particular, it has been demonstrated
mathematically that the error rate always tends to converge as the number of trees rises. In general, because
they can quickly adjust to nonlinearities detected in the data, random decision forests tend to predict better
than linear regression [30]. Random forest regression produces better results compared to other algorithms,
such as support vector machines and Neural Networks, and it is also robust against over lifting [31]. This
algorithm can forecast the result by running an unpruned regression on each n-ary tree from the training data
and then combining the results of the nary tree forecasts.
1280 Kruti Lavingia, Raj Patel, Vivek Patel, Ami Lavingia

3.4. Decision Tree Regression. A tree-based structure called decision tree regression is used to forecast
the dependent variable’s numerical results. Quinlan’s M5 algorithm [32] is implemented using what is known
as the M5P algorithm. M5P is a tree-based structure similar to CART (Classification And Regression Trees).
M5P-based trees have a multivariate linear model, whereas regression trees had values at the leaves. The trees
generated by the classification and regression trees are generally more prominent than M5P-generated model
trees.
A tree is built using the usual decision-tree method. This decision tree employs splitting criteria to account
for intra-subset variation in the class values of the samples that go down each branch. The formula below can
be used to calculate the standard deviation decrease.
X |Ti |
SDR = sd(T ) − sd(Ti ) (3.8)
i
|T |

The abrupt discontinuities that would inevitably arise between nearby linear models at the pruned tree’s
leaves are corrected by using a method.
After the machine learning algorithms had been applied to the necessary datasets, five key statistical
indicators were used as performance and assessment criteria to evaluate the success of the algorithms. Indices
[33] are Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Relative Absolute Error(RAE), Root
Relative Squared Error (RRSE), and Correlation Coefficient R2 .
In essence, they are utilizing the ML algorithm to calculate the error between predicted effort and actual
effort found in the dataset. Assuming that à is the real effort (dependent variable, to be predicted), Ā is the
mean of A, and n is the number of individual data points available. The following formulae can be used to
compute the error measures. Equations for the indices are:
n
1X
M AE = |Ai − Ãi | (3.9)
n i=1

v
u n
u1 X
RM SE = t (Ai − Ãi )2 (3.10)
n i=1

Pn
|Ai − Ãi |
RAE = Pni=1 (3.11)
n=1 |Ai − Āi |

sP
n
|Ai − Ãi |
RRAE = Pni=1 (3.12)
n=1 |Ai − Āi |

Pn
2 (Ai − Ãi )2
R = 1 − Pni=1 2
(3.13)
n=1 (Ai − Āi )

4. Research Methodology.
4.1. Data Pre Processing. The data utilized in this work is from a NASA software engineering dataset
(NASA93) that is publicly available and contains information on several projects that have undergone devel-
opment at NASA throughout the years. The dataset is structured in ARFF (Attribute Relation File Format)
format and comprises 93 rows and 26 columns. The nominal values have been transformed into comparable
values for better training. After the descriptive columns that didn’t help with training were removed, they were
discarded. To make the dataset easier to work with throughout the computation, it is then transformed into
the comma-separated values format. Table 4.1 shows a detailed description of the data being used to train the
various models
 Software Effort Estimation using Machine Learning Algorithms 1281

Table 4.1: Description of the data

Attribute Symbol Description Datatype

Project project name Name String
Category of application cat2 Which field is this project re- String
lated to
System forg Flight system or Ground sys- character [ f , g ]
tem
NASA center center Which NASA center had Number between 1 to 6
worked on the project
Capability of analyst acap
Capability of programmers pcap
Domain experience aexp
Current programming tech- modp
Increase these to decrease effort Positive integer
niques
Software tool usage tool
Experience with programming lexp
languages
Experience with VM vexp
Time restriction sced
The primary memory restric- stor
tion
Database size data
Runtime restriction on CPU time Increase these to decrease effort Positive integer
Turnaround time turn
Machine volatility virt
The difficulty of the process cplx
Required Software reliability rely
Equivalent physical line of code equivphyskloc Kilo lines of code Positive integer
Development effort act_effort The effort in terms of person- Positive integer
month

Table 4.2: Parameters By GridSearchCV for SVR

Variable Value
c 10
Gamma Auto
Kernel Linear

4.2. Model Training. “Development effort” is the primary dependent variable that is under study. The
dataset’s actual value will be compared to the expected value as part of the prediction process carried out by
the Machine Learning algorithms. The given dataset predominantly consists of more than 70% numerical data.
The data set is then divided at random into a training set and a testing set at a ratio of 70:30.

4.3. Evaluation, Results, and Discussion. In this section, the result of the experiment is discussed
and displayed. The default parameters were employed for the LinearRegression and DecisionTreeRegression
models. The parameters for SupportVectorRegression are displayed in the table 4.2. In the case of Random-
ForestRegression, all default parameters are utilized except for the specification that sets max_depth to 4. In
table 4.2, Best parameters returned by GridSearchCV algorithm for SVR is shown.
Figures 4.1 to 4.3 shows the relationship between various independent variable and dependent variable.
Figure 4.4 provides a clear illustration of the relationship between CPU runtime restrictions and both
database size and memory requirements. As the database size and memory requirements expand, the CPU
1282 Kruti Lavingia, Raj Patel, Vivek Patel, Ami Lavingia

Fig. 4.1: Plotting Dependent Variable(Effort) vs. Independent Variables

Fig. 4.2: Plotting Dependent Variable(Effort) vs. Independent Variables

experiences increased runtime restrictions due to the necessity for it to handle a greater volume of tasks within
the same time frame. Furthermore, an interesting correlation emerges: an escalation in CPU runtime restrictions
corresponds to a concurrent increase in the required software reliability. This suggests a dependency between
these variables.
Additionally, it is noteworthy that as an analyst’s capability improves, there is a parallel enhancement in
their domain familiarity. This underscores the positive relationship between analyst competency and domain
knowledge acquisition.
The outcome obtained using various error-measuring methods on various models is shown in table 4.3
demonstrates that the Random Forest method’s anticipated value and the actual value have a very close
connection. Usually R2 values are between −1 and 1. For highly correlated data R2 value is closer to 1.
Random Forest has the highest correlation coefficient compared to all other models. The other techniques
for model comparison include MAE and RAE%, RMSE and RRSE%. RMSE is commonly utilized and is
regarded as a preferable all-purpose error measure for numerical forecasts. In general, random decision forests
tend to predict better than linear regression because they can quickly adjust to nonlinearities detected in the
data. Since the Root Mean Squared Error value for Support Vector Regression is smaller, it suggests that its
predictions on the test data are more accurate compared to the other models used in this study.
4.4. Conclusion. To establish the cost required staff, and schedule for software development, it is the
job of the project manager to estimate the effort of development. The findings of several studies indicate that
early-stage project estimating errors is the primary cause of software project failures. The performance of any
estimating approach depends on a number of factors, including project complexity, project duration, personnel
skill, development process, and others. The usage of several cost-estimating methodologies is reviewed in this
essay. This paper’s contribution is the improvement of our understanding of the subject of research provided
by the literature review. No approach, namely along the RMSE dimension, estimates software development
effort especially well in the absolute sense but comparing relatively, Random Forest Regression yields the best
results as it has the lowest R2 and MAE. Additionally, practical applications of ML-based approaches could
 Software Effort Estimation using Machine Learning Algorithms 1283

Fig. 4.3: Plotting Dependent Variable(Effort) vs. Independent Variables

Fig. 4.4: Heat map showing Correlation between different Independent variables

involve project managers and data scientists working collaboratively to gather relevant data, including historical
project data, developer expertise, project complexity, and duration. By employing machine learning algorithms
on this data, project managers can enhance their ability to make more accurate early-stage project estimates,
potentially reducing the risk of project failures and improving overall project management. Future research
should focus on selecting optimal metrics for cost assessment, particularly leveraging computational intelligence
methods.

REFERENCES

[1] Dhamija, A. & Sikka, S. A Review Paper on Software Engineering Areas Implementing Data Mining Tools & Techniques.
International Journal Of Computational Intelligence Research (IJCIR). 13 pp. 559-574 (2017,5)
1284 Kruti Lavingia, Raj Patel, Vivek Patel, Ami Lavingia

Table 4.3: Outcomes of ML models using error measuring indices

Method Name MAE RMSE RAE RRSE R2

Linear Regression 760.11 1365.5 1.45 1.24 -0.41
Support Vector Regression 178.7 323.61 1.05 1.042 -0.067
Random Forest Regression 642.5 1481.7 0.85 1.070 0.2781
Decision Tree Regression 760.0 1666.1 0.82 1.098 0.0873

[2] Srinivasan, K. & Fisher, D. Machine learning approaches to estimating software development effort. IEEE Transactions On
Software Engineering. 21, 126-137 (1995)
[3] Sharma, P. & Singh, J. Systematic Literature Review on Software Effort Estimation Using Machine Learning Approaches.
2017 International Conference On Next Generation Computing And Information Systems (ICNGCIS). pp. 43-47 (2017)
[4] Moloekken-OEstvold, K., Joergensen, M., Tanilkan, S., Gallis, H., Lien, A. & Hove, S. A survey on software estimation in
the Norwegian industry. 10th International Symposium On Software Metrics, 2004. Proceedings.. pp. 208-219 (2004)
[5] Dakwala, A. & Lavingia, K. A Machine learning approach to improve the efficiency of Fake websites detection Techniques.
[6] Walkerden, F. & Jeffery, R. An Empirical Study of Analogy-based Software Effort Estimation. Empirical Software Engineering.
4, 135-158 (1999,6,1), https://doi.org/10.1023/A:1009872202035
[7] Hughes, R. Expert judgement as an estimating method. Information And Software Technology. 38, 67-75 (1996),
https://www.sciencedirect.com/science/article/pii/0950584995010459
[8] Shepperd, M. & Schofield, C. Estimating software project effort using analogies. IEEE Transactions On Software Engineering.
23, 736-743 (1997)
[9] Royce, W. Managing the development of large software systems: concepts and techniques. Proceedings Of The 9th Interna-
tional Conference On Software Engineering. pp. 328-338 (1987)
[10] Baker, F. & Kim, S. Item response theory: Parameter estimation techniques. (CRC press,2004)
[11] McManus12, J. & Wood-Harper, T. Understanding the sources of information systems project failure. (2007)
[12] Ami, R., Mehta, V. & Lavingia, K. Analyzing the non-linear effects in DWDM optical network using MDRZ modulation
format. International Journal Of Advance Engineering And Research Development (IJAERD) E-ISSN. pp. 2348-4470
[13] Boehm, B., Abts, C., Brown, A., Chulani, S., Clark, B., Horowitz, E., Madachy, R., Reifer, D. & Steece, B. Software cost
estimation with COCOMO II. (Prentice Hall Press,2009)
[14] Manikavelan, D. & Ponnusamy, R. Software quality analysis based on cost and error using fuzzy combined COCOMO model.
Journal Of Ambient Intelligence And Humanized Computing. pp. 1-11 (2020)
[15] Nandal, D. & Sangwan, O. Software cost estimation by optimizing COCOMO model using hybrid BATGSA algorithm.
International Journal Of Intelligent Engineering And Systems. 11, 250-263 (2018)
[16] Shukla, S. & Kumar, S. Applicability of neural network based models for software effort estimation. 2019 IEEE World
Congress On Services (SERVICES). 2642 pp. 339-342 (2019)
[17] Lavingia, K. & Mehta, R. Information retrieval and data analytics in internet of things: current perspective, applications and
challenges. Scalable Computing: Practice And Experience. 23, 23-34 (2022)
[18] Sarro, F., Petrozziello, A. & Harman, M. Multi-objective software effort estimation. 2016 IEEE/ACM 38th International
Conference On Software Engineering (ICSE). pp. 619-630 (2016)
[19] BaniMustafa, A. Predicting Software Effort Estimation Using Machine Learning Techniques. 2018 8th International Confer-
ence On Computer Science And Information Technology (CSIT). pp. 249-256 (2018)
[20] Anandhi, V. & Chezian, R. Regression techniques in software effort estimation using cocomo dataset. 2014 International
Conference On Intelligent Computing Applications. pp. 353-357 (2014)
[21] Yeh, T. & Deng, S. Application of machine learning methods to cost estimation of product life cycle. International Journal
Of Computer Integrated Manufacturing. 25, 340-352 (2012)
[22] Osareh, A. & Shadgar, B. Machine learning techniques to diagnose breast cancer. 2010 5th International Symposium On
Health Informatics And Bioinformatics. pp. 114-120 (2010)
[23] Rahikkala, J., Hyrynsalmi, S., Leppänen, V. & Porres, I. The role of organisational phenomena in software cost estimation:
A case study of supporting and hindering factors. E-Informatica Software Engineering Journal. 12, 167-198 (2018)
[24] Rong, S. & Bao-Wen, Z. The research of regression model in machine learning field. MATEC Web Of Conferences. 176 pp.
01033 (2018)
[25] Drucker, H., Burges, C., Kaufman, L., Smola, A. & Vapnik, V. Support vector regression machines. Advances In Neural
Information Processing Systems. 9 (1996)
[26] Xu, S., An, X., Qiao, X., Zhu, L. & Li, L. Multi-output least-squares support vector regression machines. Pattern Recognition
Letters. 34, 1078-1084 (2013), https://www.sciencedirect.com/science/article/pii/S0167865513000196
[27] Joy, R., Lavingia, A. & Lavingia, K. Performance evaluation of transmission distance and bit rates in inter-satellite optical
wireless communication system. Int J Adv Technol Eng Sci (IJATES). 4, 36-40 (2021)
[28] Zhong, H., Wang, J., Jia, H., Mu, Y. & Lv, S. Vector field-based support vector regression for building energy consumption pre-
diction. Applied Energy. 242 pp. 403-414 (2019), https://www.sciencedirect.com/science/article/pii/S0306261919304878
[29] Breiman, L. Random forests. Machine Learning. 45, 5-32 (2001)
[30] Schonlau, M. & Zou, R. The random forest algorithm for statistical learning. The Stata Journal. 20, 3-29 (2020) (2019)
 Software Effort Estimation using Machine Learning Algorithms 1285

[31] Ho, T. Random decision forests. Proceedings Of 3rd International Conference On Document Analysis And Recognition. 1 pp.
278-282 (1995)
[32] Quinlan, J. & Others Learning with continuous classes. 5th Australian Joint Conference On Artificial Intelligence. 92 pp.
343-348 (1992)
[33] Al Asheeri, M. & Hammad, M. Machine learning models for software cost estimation. 2019 International Conference On
Innovation And Intelligence For Informatics, Computing, And Technologies (3ICT). pp. 1-6

Edited by: Katarzyna Wasielewska-Michniewska

Regular paper
Received: Apr 1, 2023
Accepted: Jan 8, 2024
Copyright of Scalable Computing: Practice & Experience is the property of Scalable
Computing: Practice & Experience and its content may not be copied or emailed to multiple
sites or posted to a listserv without the copyright holder's express written permission.
However, users may print, download, or email articles for individual use.