The increasing complexity of software development projects necessitates advanced tools and methodologies for effective project management. Predictive analytics, specifically the estimation of issue resolution times, plays a crucial role in enhancing decision-making processes, resource allocation, and project planning. This paper presents an in-depth analysis of various machine learning models applied to a meticulously prepared dataset Agile Effort Estimation Software dataset (AgES) derived from GitHub public repositories. The dataset includes a comprehensive range of features such as contributors’ expertise, issue categories, and components, which are pivotal for predicting the time taken to resolve issues. Utilizing a mix of traditional and advanced machine learning techniques, including neural networks, random forests, and linear regression models, this study aims to determine the most effective methodologies for predicting issue resolution times. Each model’s performance is meticulously evaluated using a range of metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE) and Median Absolute Error (MdAE). The evaluation not only underscores the predictive accuracy of the models but also highlights their practical implications in real-world software project management scenarios. The findings of this research are intended to aid project managers and developers by providing a predictive tool that can forecast issue resolution timelines with high accuracy, thereby enabling better planning and resource management. This paper contributes to the field by detailing the model training processes, feature importance, and the potential of machine learning in transforming project management practices in the software development industry. The comparative analysis is done among the various ML techniques for the AgES dataset. The results obtained are also compared against the existing Agile datasets TAWOS and Choet et al. to show the performance of the AgES dataset. The experimental results show that the XGBoost algorithm generally performs best on the considered error metrics across the datasets.
Shankar et al. (Wed,) studied this question.