AI-Based Approaches for Software Tasks Effort Estimation: A

Systematic Review of Methods and Trends

Bruno Budel Rossi

and Lisandra Manzoni Fontoura

Programa de P

os-Graduac¸

ao em Ci

encia da Computac¸

ao, Federal University of Santa Maria (UFSM), Brazil

{bbrossi, lisandra}@inf.ufsm.br

Keywords:

Task Effort Estimation, Machine Learning, Neural Networks, Large Language Models, Natural Language

Processing.

Abstract:

Accurate measurement of task effort in software projects is essential for effective management and project

success in software engineering. Conventional methods often face limitations in both accuracy and their ability

to adapt to the complexities of contemporary projects. This systematic analysis examines the use of ensemble

learning methods and other artiﬁcial intelligence strategies for estimating task effort in software projects.

The review focuses on methods that employ machine learning, neural networks, large language models, and

natural language processing to improve the accuracy of effort estimation. The use of expert opinion is also

discussed, along with the metrics utilized in task effort estimation. A total of 826 empirical and theoretical

studies were analyzed using a comprehensive search across the ACM Digital Library, IEEE Digital Library,

ScienceDirect, and Scopus databases, with 66 studies selected for further analysis. The results highlight

the effectiveness, current trends, and beneﬁts of these techniques, suggesting that adopting AI could lead to

substantial improvements in effort estimation accuracy and more efﬁcient software project management.

1 INTRODUCTION

Accurate task effort estimation is vital for success-

ful software project management, as it directly in-

ﬂuences scheduling, resource allocation, and project

outcomes. Unlike general project effort estimation,

which evaluates the total effort for an entire project,

task effort estimation predicts the effort for individual

tasks or subtasks. This detailed approach is crucial in

agile and iterative environments where tasks are fre-

quently reassigned and reprioritized (Ali and Gravino,

2019).

Traditional methods like COCOMO-II (Boehm

et al., 2000), Use Case Points (UCP) (Karner, 1993),

and Function Point Analysis (FPA) (Albrecht and

Gaffney, 1983) often lack accuracy and adaptability

at the task level, leading to delays and cost overruns.

These approaches aggregate effort estimates broadly,

overlooking the nuances and variability speciﬁc to in-

dividual task estimations.

Advancements in AI have introduced ensemble

learning, which combines multiple machine learn-

ing models to enhance prediction robustness (Rahman

https://orcid.org/0009-0000-8153-1564

https://orcid.org/0000-0002-4669-1383

et al., 2024). Techniques such as artiﬁcial neural net-

works (ANNs) and hybrid models integrating differ-

ent algorithms have proven effective in improving the

accuracy of task-level estimations (Bilgaiyan et al.,

2019; Kumar and Srinivas, 2023). Additionally, deep

learning models like LSTM and GRU have been ex-

plored for their ability to model sequential data and

human factors in tasks, capturing the complexities of

task effort estimation (Iordan, 2024).

This systematic review synthesizes recent liter-

ature on AI techniques speciﬁcally applied to task

effort estimation in software projects. Our review

stands out by exploring not only the techniques used

to predict effort in software project tasks but also the

role of expert opinion in this process. In addition,

we provide a detailed analysis of the metrics used

to evaluate model accuracy to provide a comprehen-

sive overview of current practices and their efﬁcacy in

task-level estimation.

Our study provides an up-to-date overview of ef-

fort estimation practices, emphasizing task-focused

approaches. Examining the strengths and limitations

of AI-based techniques offers valuable insights to ad-

vance research and support adopting more accurate

methods in software engineering.

The paper is organized as follows. Section 2 re-

144

Rossi, B. B. and Fontoura, L. M.

AI-Based Approaches for Software Tasks Effort Estimation: A Systematic Review of Methods and Trends.

DOI: 10.5220/0013218200003929

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 27th International Conference on Enterprise Information Systems (ICEIS 2025) - Volume 2, pages 144-151

ISBN: 978-989-758-749-8; ISSN: 2184-4992

views related work, Section 3 describes our method-

ology, Section 4 discusses the results, and Section 5

concludes the article with future directions.

2 RELATED WORKS

Several systematic reviews, such as Cabral et al.

(2023), have explored Ensemble Effort Estimation

(EEE), focusing on dynamic ensemble selection tech-

niques. Our study speciﬁcally addresses the use of

ensemble methods for task effort estimation, differ-

ing from broader reviews like Ali and Gravino (2019),

which analyze ML techniques, including artiﬁcial

neural networks (ANN) and support vector machines

(SVM), across various contexts. Unlike Brar and

Nandal (2022), who focus on ML techniques for gen-

eral effort estimation, we concentrate on task-level es-

timations, emphasizing the unique challenges and ad-

vantages of ensemble models.

Usman et al. (2014) focus on effort estimation in

agile development, using expert judgment, Planning

Poker, and AI-based methods. Their focus on ag-

ile environments contrasts with our broader approach,

which encompasses diverse software project contexts.

Other notable studies include Dehghan et al.

(2016), who proposed a hybrid model combining tex-

tual techniques and task metadata, and Bilgaiyan et al.

(2019), who investigated neural networks in agile en-

vironments.

These studies provide a foundation for under-

standing the application of ML and AI in software

effort estimation. Our research stands out by focus-

ing on ensemble models and exploring their potential

to enhance the accuracy and efﬁciency of task effort

estimation.

3 RESEARCH METHOD

This systematic review followed the guidelines pro-

posed by Kitchenham (2004) and Kitchenham et al.

(2009) to ensure comprehensive and unbiased cov-

erage of the relevant literature, aiming to provide a

complete overview of the use of ensemble models and

other AI techniques for task effort estimation in soft-

ware projects.

The Parsifal tool (https://parsif.al) was used to as-

sist in the search and selection of studies, ensuring a

rigorous and replicable methodology. The research

was conducted in the ACM Digital Library, IEEE

Digital Library, ScienceDirect, and Scopus databases,

which are highly relevant sources for the ﬁeld.

3.1 Planning

The PICOC technique was employed to deﬁne the

scope of this systematic review, guiding the formu-

lation of research questions and the search for rele-

vant evidence. This approach structured the review

by focusing on the population (software projects and

development teams), intervention (AI techniques like

machine learning and ensemble models), comparison

(against traditional methods such as COCOMO and

function point analysis), outcome (improvements in

estimation accuracy and efﬁciency), and context (soft-

ware development environments). PICOC ensured a

comprehensive and well-deﬁned framework for ana-

lyzing the impact of AI-based techniques on task ef-

fort estimation.

3.1.1 Research Questions

The following research questions guided the data ex-

traction and analysis:

RQ1: Which AI-based techniques are most com-

monly used for task effort estimation in software

projects, and why are they chosen?

RQ2: Are experts still necessary for task effort es-

timation or for training AI models? If so, how are

they employed?

RQ3: What metrics or benchmarks are commonly

used to evaluate the performance of ensemble learn-

ing models in task effort estimation?

These questions aim to clarify the current land-

scape of AI use in effort estimation and to identify

best practices and gaps in the literature.

3.1.2 Search Strategy and Data Sources

To ensure the retrieval of relevant articles, we em-

ployed the following search string: (“software de-

velopment” OR “software project”) AND (“ensem-

ble learning” OR “artiﬁcial intelligence” OR “neu-

ral network” OR “deep learning” OR “large language

model” OR “machine learning”) AND (“effort esti-

mation”) AND year ≥ 2014.

The review used widely recognized scientiﬁc

databases to ensure the relevance and quality of the

included studies. The selected databases were ACM

Digital Library, IEEE Digital Library, ScienceDirect,

and Scopus, providing a solid foundation for the sys-

tematic review.

3.2 Conducting

Figure 1 summarizes the process of selecting the ar-

ticles, outlining the three steps taken and the number

of articles analyzed and selected at each step.

AI-Based Approaches for Software Tasks Effort Estimation: A Systematic Review of Methods and Trends

145

Figure 1: Article search protocol.

After eliminating duplicates using the Parsifal

tool, 198 articles were removed, leaving 638 for de-

tailed analysis. In the second step, based on the read-

ing of the title and keywords, the inclusion and exclu-

sion criteria were applied, resulting in 328 selected

articles (see Section 3.2.1). In the ﬁnal selection step,

the articles were read in full, resulting in 66 articles

being selected for analysis (see Section 3.2.2). Efforts

included using institutional subscriptions and contact-

ing authors to access publications, although four ar-

ticles were excluded due to inaccessibility, ensuring

transparency in the process.

3.2.1 Inclusion and Exclusion Criteria

The selection criteria ensured the quality and rele-

vance of the articles:

Inclusion Criteria (IC). Studies published between

2014 and 2024 applying AI techniques to effort esti-

mation in software tasks and empirical or theoretical

articles explicitly addressing AI applications in effort

estimation.

Exclusion Criteria (EC). Studies that do not address

effort estimation or software development, those that

do not use AI, publications before 2014, and inacces-

sible articles.

After applying these criteria, 328 articles were se-

lected for the quality assessment phase, focusing on

their title and keywords to evaluate their contributions

to effort estimation in software projects using AI and

ensemble models.

3.2.2 Quality Assessment Checklist

To ensure the quality of the included studies, we ap-

plied the following questions:

Title. Is the title related to the review theme?

Keywords. Are the keywords relevant to the theme?

Abstract. Does the abstract clearly present the

study’s objective, development, and results?

Context. Is the study set in the context of task effort

estimation using AI techniques?

Each question was scored: Yes (5 points), Par-

tially (3 points), or No (0 points). Studies scoring at

least 14 points were included in the review. After this

phase, 66 studies were selected, representing the com-

plete set used to evaluate effort estimation methods.

4 RESULTS AND DISCUSSION

Section 4.1 analyzes AI-based techniques like Ma-

chine Learning, Large Language Models, NLP, and

Neural Networks, highlighting their application and

selection for improving effort estimation (RQ1). Sec-

tion 4.2 discusses the continued role of experts in ad-

justing or validating AI predictions, particularly in

complex or data-scarce contexts (RQ2). Finally, Sec-

tion 4.3 explores metrics and benchmarks for evalu-

ating model effectiveness, emphasizing practices for

accurate predictions (RQ3).

4.1 AI-Based Techniques for Task

Effort Estimation

To address RQ1, this section explores commonly

used AI-based techniques for task effort estimation

in software projects. The studies emphasize the fre-

quent use of hybrid models combining various ma-

chine learning approaches, including neural networks

(NN) and natural language processing (NLP), lever-

aging their strengths for more accurate task effort es-

timation. These models overcome individual limita-

tions, providing robust, adaptable solutions for the

complex realities of modern software projects, where

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

146

a single method often falls short of delivering accurate

estimations.

4.1.1 Artiﬁcial Intelligence and Machine

Learning

AI and ML are signiﬁcant for effort estimation in

software projects, with distinct approaches impact-

ing estimation accuracy. AI without ML involves

techniques like rule-based systems and Bayesian net-

works, effective in modeling uncertainties and inter-

dependent variables. Dragicevic et al. (2017) show-

cased Bayesian networks’ ability to handle incom-

plete data and probabilistic inferences, making them

suitable for uncertain data scenarios.

However, AI without ML lacks dynamic learn-

ing, limiting its use in evolving data contexts, such as

modern software projects. Bayesian networks model

agile tasks effectively but struggle with new, unantic-

ipated inputs during initial modeling phases.

Conversely, ML is a widely used approach, with

62 of the 66 reviewed articles applying techniques

like SVM, KNN, and decision trees to enhance esti-

mation accuracy. Studies like Ali and Gravino (2019)

highlight the superior performance of ANN and SVM

in modeling complex nonlinear relationships and han-

dling smaller, noisier datasets.

A notable trend is combining AI and ML to lever-

age their strengths. Bilgaiyan et al. (2019) explored

hybrid models combining feedforward neural net-

works with rule-based models, achieving greater ac-

curacy in various software development stages. Tech-

niques like ensemble learning, as demonstrated by

Azzeh et al. (2018), further improve estimates by re-

ducing bias across multiple ML models.

The reviewed studies indicate that ML methods,

particularly when integrated with other techniques,

generally produce better results than purely AI-based

methods. Neural networks and SVM remain highly

effective, while hybrid models such as Bayesian net-

works and ensemble learning are emerging as promis-

ing solutions for accurate effort estimation.

These evolving approaches in effort estimation of-

fer robust solutions for handling the complexity and

variability of modern software projects. As Sarro

et al. (2022) noted, the integration of AI and ML en-

ables continuous improvement in predictions, over-

coming traditional methods’ limitations.

4.1.2 Large Language Models

Alhamed and Storer (2022) explore the use of LLMs,

such as GPT, in effort estimation for software main-

tenance, demonstrating their efﬁciency in processing

large volumes of textual data. They propose a hybrid

framework combining LLM predictions with human

expert insights to enhance estimation accuracy.

The study highlights that while LLMs excel in au-

tomated large-scale data analysis, human experts are

crucial for interpreting contextual nuances, especially

in complex, dynamic maintenance tasks. This hybrid

approach signiﬁcantly improves estimation accuracy

and adaptability, as conﬁrmed by practical case stud-

ies across various maintenance scenarios.

The integration of LLMs with human expertise re-

duces errors in high variability contexts where tra-

ditional methods fall short. The study underscores

the evolving role of LLMs in software engineering,

emphasizing the balance between automation and hu-

man intervention for more reliable and nuanced pre-

dictions.

4.1.3 Natural Language Processing

The use of NLP techniques in task effort estimation

for software projects has become prominent, with

nine of the 66 reviewed articles employing various

NLP approaches to improve accuracy. Techniques

like topic modeling, as used by Yasmin (2024) with

Latent Dirichlet Allocation (LDA), help reduce tex-

tual complexity by focusing on relevant topics. How-

ever, LDA may struggle with capturing nuanced de-

tails of complex tasks, leading to less accurate esti-

mates in certain contexts.

Semantic embedding techniques, a subset of

machine learning approaches like SBERT and

Word2Vec, have also been widely applied. Yalc¸iner

et al. (2024) demonstrate SBERT’s capability to gen-

erate rich semantic representations, enhancing task

description comparisons despite its high computa-

tional demands. Conversely, Dan et al. (2024) high-

light Word2Vec’s efﬁciency in identifying task simi-

larities with lower computational costs, though it may

be less effective in understanding complex sentences

compared to SBERT.

Hybrid models combining NLP with other tech-

niques, as explored by Dehghan et al. (2016), inte-

grate NLP methods with ensemble learning and task

metadata to provide comprehensive task views and

robust estimates. Despite their effectiveness, these

models can increase computational complexity and

reduce result explainability. NLP’s ability to process

unstructured data, like natural language task descrip-

tions, is invaluable, but challenges like the need for

high-quality training data and the interpretability of

complex model results persist.

AI-Based Approaches for Software Tasks Effort Estimation: A Systematic Review of Methods and Trends

147

4.1.4 Neural Networks

Neural networks are widely adopted in effort estima-

tion for software projects, with 44 reviewed articles

utilizing these techniques. Their strength lies in cap-

turing the complexity of project data, modeling non-

linear relationships, and identifying hidden patterns

often missed by traditional methods. However, the

high training costs in terms of time and computational

resources are notable challenges. Commonly used

techniques include Deep Neural Networks (DNNs),

Convolutional Neural Networks (CNNs), Recurrent

Neural Networks (RNNs), including Long Short-

Term Memory (LSTM) networks, and Cascade Cor-

relation Neural Networks (CCNN), each with speciﬁc

strengths and limitations.

DNNs are praised for their predictive accuracy

with extensive data, but require considerable tuning

and large volumes of training data, as shown by Ali

and Gravino (2019). Radial Basis Function Neu-

ral Networks (RBFNNs), discussed by Nassif et al.

(2016), offer faster training and are effective for well-

deﬁned clusters, though they may not perform as well

with complex tasks. Cascade Correlation Neural Net-

works provide adaptability by adjusting their structure

during training, reducing training time, and showing

lower overﬁtting tendencies.

RNNs, particularly LSTM networks, are effective

in modeling sequential data and capturing long-term

dependencies, crucial for evolving projects. Iordan

(2024) demonstrated LSTM’s superior performance

when optimized with particle swarm optimization.

Despite their effectiveness, RNNs and LSTMs de-

mand signiﬁcant computational power and face chal-

lenges like the vanishing gradient problem. CNNs,

though traditionally used in image processing, have

been adapted for time series and textual data, proving

useful in speciﬁc scenarios but less common in effort

estimation compared to RNNs and LSTMs.

Overall, neural networks are highly adaptable and

effective for handling large volumes of unstructured

data, capturing complex patterns in software projects

of varying scope and complexity. However, their

training is resource-intensive, and the ”black box” na-

ture of deeper networks complicates result interpreta-

tion, posing challenges for project managers who re-

quire transparency in predictive models.

4.2 Use of Expert Opinion

To address RQ2, twelve articles incorporating ex-

pert opinions were selected, highlighting their role in

hybrid systems for model validation or adjustment.

While ML algorithms excel in identifying patterns

from large datasets, they often lack the context needed

to account for human or circumstantial factors in ef-

fort estimation. Alhamed and Storer (2022) propose

a hybrid framework combining language models and

expert opinion, enhancing estimation accuracy. Sim-

ilarly, Meiliana et al. (2023) and Dan et al. (2024) il-

lustrate the integration of expert input to calibrate and

adjust automated models, ensuring predictions align

with project-speciﬁc nuances.

Despite advancements in automation, expert re-

liance remains signiﬁcant, particularly in complex or

early-stage projects where historical data may be in-

sufﬁcient for AI models. Experts provide critical con-

textual insights that automated systems often miss.

Yasmin (2024) demonstrate the importance of expert

adjustment in cases of signiﬁcant project variations

or ambiguous AI model outputs. This expert inter-

vention bridges gaps in automated models, especially

when project variables deviate from historical norms.

The main advantage of expert involvement is their

ability to contextualize predictions based on emerg-

ing variables, enhancing estimation accuracy. How-

ever, it can also introduce subjectivity and potential

delays in decision-making, as discussed by Meiliana

et al. (2023). Despite ongoing advancements in AI,

expert opinion remains crucial, particularly in scenar-

ios where data scarcity or complexity challenges AI

efﬁcacy. The trend is toward integrating expert ad-

justments with automated predictions to enhance ac-

curacy and reliability, though reliance on experts is

expected to decrease as AI models evolve.

4.3 Metrics Used to Evaluate Model

Performance

This section covers metrics commonly used to eval-

uate model performance in task effort estimation

(RQ3).

Mean Magnitude of Relative Error (MMRE).

Highlighted by Bilgaiyan et al. (2019) and Satapathy

et al. (2014), MMRE measures the difference between

predicted and actual effort.

Median Magnitude of Relative Error (MdMRE).

Used alongside MMRE, it mitigates the impact of out-

liers, as noted by Zakrani et al. (2019).

PRED(X): Evaluates the percentage of predictions

within a given error range, typically 25%, as seen in

Sarro et al. (2022).

Mean Absolute Error (MAE) and Root Mean

Square Error (RMSE). These metrics measure the

mean absolute and squared errors, sensitive to huge

prediction errors, frequently used in Phan and Jan-

nesari (2022) for neural network performance anal-

ysis.

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

148

Comparing traditional models with AI and ML

benchmarks, such as neural networks and ensemble

learning, is crucial for assessing improvements in esti-

mation accuracy over conventional methods like CO-

COMO II and UCP (Cabral et al., 2023).

5 CONCLUSIONS

The application of AI techniques, particularly ML,

has become a dominant approach for effort estimation

in software projects, as evidenced by the majority of

the 66 analyzed studies. ML’s versatility and effec-

tiveness in handling complex data make it a preferred

choice among researchers.

Conversely, the adoption of AI models without

ML, such as Bayesian networks, remains limited, and

LLM usage is still in its early stages. A signiﬁcant

trend observed is the integration of hybrid models,

which combine various techniques to produce more

robust and accurate results.

Expert opinions remain to play a crucial role in

certain studies, particularly for calibrating and vali-

dating estimates in complex scenarios. The integra-

tion of human expertise with AI models ensures that

predictions are tailored to the speciﬁc nuances of each

project, enhancing accuracy.

Evaluation metrics such as MMRE, MdMRE, and

PRED(X) are critical for assessing model accuracy

and verifying improvements over traditional methods.

These metrics are instrumental in demonstrating the

effectiveness of AI-based models in improving effort

estimation accuracy in software projects.

ACKNOWLEDGEMENTS

We thank the Brazilian Army and its Army Strategic

Program ASTROS for the ﬁnancial support through

the SIS-ASTROS GMF project (898347/2020).

REFERENCES

Abdelali, Z., Mustapha, H., and Abdelwahed, N. (2019). In-

vestigating the use of random forest in software effort

estimation. In Procedia Computer Science, volume

148, pages 343–352. Elsevier B.V.

Ahmad, F. B. and Ibrahim, L. M. (2022). Software ef-

fort estimation based on long short term memory and

stacked long short term memory. In 2022 8th In-

ternational Conference on Contemporary Information

Technology and Mathematics, ICCITM 2022, pages

165–170. Institute of Electrical and Electronics Engi-

neers Inc.

Albrecht, A. J. and Gaffney, J. E. (1983). Software func-

tion, source lines of code, and development effort pre-

diction: A software science validation. IEEE Trans-

actions on Software Engineering, (6):639–648.

Alhamed, M. and Storer, T. (2022). Evaluation of context-

aware language models and experts for effort esti-

mation of software maintenance issues. In Proceed-

ings - 2022 IEEE International Conference on Soft-

ware Maintenance and Evolution, ICSME 2022, pages

129–138. Institute of Electrical and Electronics Engi-

neers Inc.

Ali, A. and Gravino, C. (2019). A systematic literature

review of software effort prediction using machine

learning methods.

Alsaadi, B. and Saeedi, K. (2024). Ensemble effort estima-

tion for novice agile teams. Information and Software

Technology, 170.

Arachchi, K. J. W. and Amalraj, C. R. (2023). An ag-

ile project management supporting approach for es-

timating story points in user stories. In Proceedings

of ICITR 2023 - 8th International Conference on In-

formation Technology Research: The Next Evolution

in Digital Transformation. Institute of Electrical and

Electronics Engineers Inc.

Azzeh, M., Nassif, A. B., Banitaan, S., and Lopez-Martin,

C. (2018). Ensemble of learning project productiv-

ity in software effort based on use case points. In

Proceedings - 17th IEEE International Conference on

Machine Learning and Applications, ICMLA 2018,

pages 1427–1431. Institute of Electrical and Electron-

ics Engineers Inc.

Bilgaiyan, S., Mishra, S., and Das, M. (2019). Effort es-

timation in agile software development using experi-

mental validation of neural network models. Interna-

tional Journal of Information Technology (Singapore),

11:569–573.

Boehm, B. W., Horowitz, E., Madachy, R., Reifer, D.,

Clark, B., Steece, B., Brown, A. W., Chulani, S., and

Abts, C. (2000). Software cost estimation with CO-

COMO II. Prentice Hall PTR.

Brar, P. and Nandal, D. (2022). A systematic literature

review of machine learning techniques for software

effort estimation models. In Proceedings - 2022

5th International Conference on Computational In-

telligence and Communication Technologies, CCICT

2022, pages 494–499. Institute of Electrical and Elec-

tronics Engineers Inc.

Cabral, J. T. H. A., Oliveira, A. L., and da Silva, F. Q.

(2023). Ensemble effort estimation: An updated and

extended systematic literature review. Journal of Sys-

tems and Software, 195.

Choetkiertikul, M., Dam, H. K., Tran, T., Pham, T., Ghose,

A., and Menzies, T. (2019). A deep learning model for

estimating story points. IEEE Transactions on Soft-

ware Engineering, 45:637–656.

Dan, I., Rusu, C., and Oliver, O. (2024). An nlp approach to

estimating effort in a work environment. In Proceed-

ings of the 2024 International Conference on Software

Effort Estimation, Cluj-Napoca, Romania. IEEE.

Dantas, E., Costa, A., Vinicius, M., Perkusich, M.,

AI-Based Approaches for Software Tasks Effort Estimation: A Systematic Review of Methods and Trends

149

Almeida, H., and Perkusich, A. (2019). An effort es-

timation support tool for agile software development:

An empirical evaluation. In Proceedings of the In-

ternational Conference on Software Engineering and

Knowledge Engineering, SEKE, volume 2019-July,

pages 82–87. Knowledge Systems Institute Graduate

School.

Dantas, E., Perkusich, M., Dilorenzo, E., Santos, D. F.,

Almeida, H., and Perkusich, A. (2018). Effort esti-

mation in agile software development: An updated re-

view. In International Journal of Software Engineer-

ing and Knowledge Engineering, volume 28, pages

1811–1831. World Scientiﬁc Publishing Co. Pte Ltd.

de A. Cabral, J. T. H., de A. Araujo, R., Nobrega, J. P.,

and de Oliveira, A. L. (2017). Heterogeneous ensem-

ble dynamic selection for software development effort

estimation. In 2017 IEEE 29th International Con-

ference on Tools with Artiﬁcial Intelligence (ICTAI),

pages 210–217.

Dehghan, A., Blincoe, K., and Damian, D. (2016). A hy-

brid model for task completion effort estimation. In

SWAN 2016 - Proceedings of the 2nd International

Workshop on Software Analytics, co-located with FSE

2016, pages 22–28. Association for Computing Ma-

chinery, Inc.

Dragicevic, S., Celar, S., and Turic, M. (2017). Bayesian

network model for task effort estimation in agile soft-

ware development. Journal of Systems and Software,

127:109–119.

Goto, T. (2016). ACIT-CSII-BCD 2016 : 4th International

Conference on Applied Computing and Information

Technology (ACIT 2016) ; 3rd International Confer-

ence on Computational Science/Intelligence and Ap-

plied Informatics (CSII 2016), 1st International Con-

ference on Big Data, Cloud Computing, Data Science

& Engineering (BCD 2016) : proceedings : 12-14 De-

cember 2016, Las Vegas, Nevada, USA. Conference

Publishing Services, IEEE Computer Society.

Gultekin, M. and Kalipsiz, O. (2020). Story point-based

effort estimation model with machine learning tech-

niques. International Journal of Software Engineering

and Knowledge Engineering, 30:43–66.

Hoda, M. N. (2023). Proceedings of the 17th INDIACom;

2023 10th International Conference on Computing for

Sustainable Global Development (INDIACom) : (15-

17 March 2023). India-Com.

Hosni, M. (2023). On the value of combiners in hetero-

geneous ensemble effort estimation. In International

Joint Conference on Knowledge Discovery, Knowl-

edge Engineering and Knowledge Management, IC3K

- Proceedings, volume 1, pages 153–163. Science and

Technology Publications, Lda.

Idri, A., Hosni, M., and Abran, A. (2016). Improved esti-

mation of software development effort using classical

and fuzzy analogy ensembles. Applied Soft Comput-

ing Journal, 49:990–1019.

Iordan, A. E. (2024). An optimized lstm neural network

for accurate estimation of software development ef-

fort. Mathematics, 12.

Iqbal, M., Ijaz, M., Mazhar, T., Shahzad, T., Abbas, Q.,

Ghadi, Y. Y., Ahmad, W., and Hamam, H. (2024). Ex-

ploring issues of story-based effort estimation in ag-

ile software development (asd). Science of Computer

Programming, 236.

Iwata, K., Nakashima, T., Anan, Y., and Ishii, N. (2017).

Machine learning classiﬁcation to effort estimation for

embedded software development projects. Interna-

tional Journal of Software Innovation, 5:19–32.

Karner, G. (1993). Metrics for object-oriented system de-

velopment. Technical report, Objectory Systems De-

veloper Report.

Kassaymeh, S., Alweshah, M., Al-Betar, M. A., Hammouri,

A. I., and Al-Ma’aitah, M. A. (2024). Software ef-

fort estimation modeling and fully connected artiﬁ-

cial neural network optimization using soft computing

techniques. Cluster Computing, 27:737–760.

Kitchenham, B., Pearl Brereton, O., Budgen, D., Turner,

M., Bailey, J., and Linkman, S. (2009). Systematic lit-

erature reviews in software engineering – a systematic

literature review. Information and Software Technol-

ogy, 51(1):7–15. Special Section - Most Cited Articles

in 2002 and Regular Research Papers.

Kitchenham, B. A. (2004). Procedures for performing sys-

tematic reviews.

Kumar, A. (2024). Recommendation of machine learning

techniques for software effort estimation using multi-

criteria decision making. Journal of Universal Com-

puter Science, 30:221–241.

Kumar, K. H. and Srinivas, K. (2023). An accurate anal-

ogy based software effort estimation using hybrid op-

timization and machine learning techniques. Multime-

dia Tools and Applications, 82:30463–30490.

Kumar, S., Arora, M., Sakshi, and Chopra, S. (2022). A

review of effort estimation in agile software devel-

opment using machine learning techniques. In 4th

International Conference on Inventive Research in

Computing Applications, ICIRCA 2022 - Proceedings,

pages 416–422. Institute of Electrical and Electronics

Engineers Inc.

Lavingia, K., Patel, R., Patel, V., and Lavingia, A. (2024).

Software effort estimation using machine learning al-

gorithms. Scalable Computing, 25:1276–1285.

Marapelli, B., Carie, A., and Islam, S. M. (2021). Software

effort estimation with use case points using ensemble

machine learning models. In International Confer-

ence on Electrical, Computer, and Energy Technolo-

gies, ICECET 2021. Institute of Electrical and Elec-

tronics Engineers Inc.

Meiliana, Daniella, G., Wijaya, N., Putra, N. G. E., and

Efata, R. (2023). Agile software development effort

estimation based on product backlog items. In Proce-

dia Computer Science, volume 227, pages 186–193.

Elsevier B.V.

Najm, A. and Zakrani, A. (2023). On the interpretability of

a tree-based ensemble for predicting software effort.

In Colloquium in Information Science and Technol-

ogy, CIST, pages 129–134. Institute of Electrical and

Electronics Engineers Inc.

Nakamura, M. (2019). Proceedings, 18th IEEE/ACIS Inter-

national Conference on Software Engineering, Artiﬁ-

cial Intelligence, Networking and Parallel/Distributed

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

150

Computing (SNPD 2019) : July 8-11, 2019, Toyama,

Japan. IEEE Computer Society.

Nassif, A. B., Azzeh, M., Capretz, L. F., and Ho, D. (2016).

Neural network models for software development ef-

fort estimation: a comparative study. Neural Comput-

ing and Applications, 27:2369–2381.

Phan, H. and Jannesari, A. (2022). Heterogeneous graph

neural networks for software effort estimation. In

International Symposium on Empirical Software En-

gineering and Measurement, pages 103–113. IEEE

Computer Society.

Potolea, R. and Slavescu, R. R. (2017). Proceedings, 2017

IEEE 13th International Conference on Intelligent

Computer Communication and Processing (ICCP) :

Cluj-Napoca, Romania, September 7-9, 2017. IEEE.

Rahman, M., Sarwar, H., Kader, A., Goncalves, T., and Tin,

T. T. (2024). Review and empirical analysis of ma-

chine learning-based software effort estimation. IEEE

Access, 12:85661–85680.

Ramessur, M. A. and Nagowah, S. D. (2021). A predic-

tive model to estimate effort in a sprint using machine

learning techniques. International Journal of Infor-

mation Technology (Singapore), 13:1101–1110.

Rankovic, N. and Rankovic, D. (2024). Power of lstm and

shap in the use case point approach for software effort

and cost estimation. In 2024 IEEE 22nd World Sympo-

sium on Applied Machine Intelligence and Informat-

ics, SAMI 2024 - Proceedings, pages 59–64. Institute

of Electrical and Electronics Engineers Inc.

Sanchez, E. R., Maceda, H. C., and Santacruz, E. V. (2022).

Software effort estimation for agile software develop-

ment using a strategy based on k-nearest neighbors al-

gorithm. In 2022 IEEE Mexican International Confer-

ence on Computer Science, ENC 2022 - Proceedings.

Institute of Electrical and Electronics Engineers Inc.

Sarro, F., Moussa, R., Petrozziello, A., and Harman, M.

(2022). Learning from mistakes: Machine learning

enhanced human expert effort estimates. IEEE Trans-

actions on Software Engineering, 48:1868–1882.

Satapathy, S. M., Acharya, B. P., and Rath, S. K. (2014).

Class point approach for software effort estimation us-

ing stochastic gradient boosting technique. ACM SIG-

SOFT Software Engineering Notes, 39:1–6.

Sharma, A. and Chaudhary, N. (2022a). Analysis of soft-

ware effort estimation based on story point and lines

of code using machine learning. International Journal

of Computing and Digital Systems, 12:131–140.

Sharma, A. and Chaudhary, N. (2022b). Prediction of soft-

ware effort by using non-linear power regression for

heterogeneous projects based on use case points and

lines of code. In Procedia Computer Science, volume

218, pages 1601–1611. Elsevier B.V.

Song, L., Minku, L. L., and Yao, X. (2018). A novel au-

tomated approach for software effort estimation based

on data augmentation. In ESEC/FSE 2018 - Proceed-

ings of the 2018 26th ACM Joint Meeting on European

Software Engineering Conference and Symposium on

the Foundations of Software Engineering, pages 468–

479. Association for Computing Machinery, Inc.

Sousa, A. O., Veloso, D. T., Goncalves, H. M., Faria, J. P.,

Mendes-Moreira, J., Graca, R., Gomes, D., Castro,

R. N., and Henriques, P. C. (2023). Applying ma-

chine learning to estimate the effort and duration of

individual tasks in software projects. IEEE Access,

11:89933–89946.

Sudarmaningtyas, P. and Mohamed, R. (2021). A review

article on software effort estimation in agile method-

ology.

Sunda, N. and Sinha, R. R. (2023). Optimizing effort es-

timation in agile software development: Traditional

vs. advanced ml methods. In 2023 International Con-

ference on Communication, Security and Artiﬁcial In-

telligence, ICCSAI 2023, pages 487–494. Institute of

Electrical and Electronics Engineers Inc.

Usman, M., Mendes, E., Weidt, F., and Britto, R. (2014).

Effort estimation in agile software development: A

systematic literature review. In ACM International

Conference Proceeding Series, pages 82–91. Associa-

tion for Computing Machinery.

Vyas, M. and Hemrajani, N. (2021). Predicting effort of

agile software projects using linear regression, ridge

regression and logistic regression. International Jour-

nal on ”Technical and Physical Problems of Engineer-

ing” (IJTPE) Issue, 47:14–19.

Yalc¸iner, B., Dinc¸er, K., Karac¸or, A. G., and Efe, M. O.

(2024). Enhancing agile story point estimation: Inte-

grating deep learning, machine learning, and natural

language processing with sbert and gradient boosted

trees. Applied Sciences, 14(16):7305.

Yasmin, A. (2024). Cost adjustment for software crowd-

sourcing tasks using ensemble effort estimation and

topic modeling. Arabian Journal for Science and En-

gineering, 49:12693–12728.

Zakrani, A., Hain, M., and Idri, A. (2019). Improving soft-

ware development effort estimation using support vec-

tor regression and feature selection. IAES Interna-

tional Journal of Artiﬁcial Intelligence, 8:399–410.

AI-Based Approaches for Software Tasks Effort Estimation: A Systematic Review of Methods and Trends

151