A Machine Learning Approach for Estimating Overtime Allocation in Software Development Projects - Random Forest Regression

Aug 3, 2026

Upload by Aria

Overtime planning in software projects has traditionally relied on search-based multiobjective optimization, but the generated solutions often fail to align with project managers’ practical intuition. This study proposes a machine learning model that learns preferred overtime allocation patterns from manager-annotated optimization solutions. Training uses 1092 instances collected from software houses, with Random Forest Regression estimating managers’ preference. Evaluation with MAE, RMSE, and R2 shows strong predictive accuracy and consistent superiority over baseline regression models, supporting reliable overtime plan estimation.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages5

Reading Time13 min

AccessFree to download

Introduction

Methodology

A Machine Learning Approach for Estimating Overtime Allocation in Software Development Projects Hammed AdeleyeMojeed Department of Computer System Architecture, Faculty of Electronics, Telecommunication and Informatics, Gdansk University of Technology Gdansk, Poland [hammed.mojeed@pg.edu.pl](hammed.mojeed@pg.edu.pl) RafalSzlapczynski Department of Applied Computer Science, Institute of Ocean Engineering and Ship Technology, Gdansk University of Technology Gdansk, Poland [rafal.szlapczynski@pg.edu.pl](rafal.szlapczynski@pg.edu.pl) Abstract Overtime planning in software projects has traditionally been approached with search-based multiobjective optimization algorithms. However, the explicit solutions produced by these algorithms often lack applicability and acceptance in the software industry due to their disregard for project managers' intuitive knowledge. This study presents a machine learning model that learns the preferred overtime allocation patterns from solutions annotated by project managers and applied to four publicly available software development projects. The model was trained using 1092 instances of annotated solutions gathered from software houses, and the Random Forest Regression (RFR) algorithm was used to estimate the PMs’ preference. The evaluation results using MAE, RMSE, and R2 revealed that RFR exhibits excellent predictive power in this domain with minimal error. RFR also outperformed the baseline regression models in all the performance measures. The proposed machine learning approach provides a reliable and effective tool for estimating project managers' preferences for overtime plans. Keywords: Software Overtime Planning, Software Project Planning, Machine Learning, Random Forest Regression. 1. Introduction Planning a software project is a complex and highly dynamic task characterized by uncertainties and the risk of overrun in duration and cost. Although PMs are provided with automated project planning tools, software development teams still suffer from unplanned overtime as the only option when the project encounters “mission creep” or an excessive change in requirements [4] . This issue of unplanned overtime has been a persistent challenge in the software industry, leading to negative impacts on developers [6], [14] and the quality of the software they build [3] . These findings have drawn researchers’ attention to more proactive overtime planning, which is the focus of our research. The current approach modeled software overtime allocation as a multi-objective optimization problem considering its effects on project duration, cost, overrun risk, and quality [2, 3], [8], leading to a new field called Software Overtime Planning (SOP) . The first search-based optimization formulation of SOP was introduced by Ferrucci et al. [3] . Subsequent studies have extended their work with multi-objective evolutionary algorithms [2], [12] and multi-objective memetic algorithms [8] to produce optimal overtime plans. Ferruci et al. [3] applied NSGA-IIv, a variant of NSGA-II specifically designed for overtime scheduling, to find the optimal overtime allocations. Similarly, Sarro et al. [12] applied Adaptivevsc, a variant ofNSGA-II that efficiently combines the crossover operator used in [3] and adaptive genetic operators to produce a dynamic strategy for selecting genetic operators as optimization progresses. Using the same NSGA-II, De Barros and De Araujo [2] incorporated the already-established effect of overtime on software quality into SOP formulation by simulating the defects introduced by developers during overtime as they affect project cost and duration. Mojeed et al. [8] introduced a memetic approach to SOP using the same experimental setting and datasets as in [2] and a multi-objective shuffled frog-leaping algorithm (MOSFLA) as the search method. These existing studies in SOP have produced quality solutions for Project Managers (PMs) to allocate overtime better. However, thes

FAQs

Why are traditional optimization solutions for software overtime planning often rejected by project managers?

They usually disregard project managers’ intuitive knowledge when producing explicit schedules, reducing their applicability and acceptance in industry practice.

What machine learning method is used to estimate project managers’ overtime preferences?

The study uses Random Forest Regression (RFR) trained on 1092 instances of project-manager annotated solutions.

How is the model evaluated and how does it perform compared with baseline models?

Performance is measured using MAE, RMSE, and R2. Results show RFR has excellent predictive power with minimal error and outperforms baseline regression models across all metrics.

Aug 3, 2026

Upload by Aria