AI-DRIVEN EFFORT AND TIME PREDICTION FOR SOFTWARE PROJECT TASKS

Abstract

Estimation of software project effort and software project time are very important in successful project planning, allocation of resources, budgeting and timely delivery. Traditional approaches to estimation are based on expert judgment and past experience, and may not be accurate due to uncertainty, evolving requirements and the lack of project information. This project aims to address these challenges by developing an AI-powered software effort and software completion time prediction framework based on advanced machine learning techniques. The proposed system uses the CESAW software project dataset and uses comprehensive data preprocessing methods such as missing value treatment, categorical encoding, feature scaling, outlier removal, and feature engineering. In order to overcome the challenges of small dataset and make the model more robust, synthetic project data is generated based on Conditional Tabular Generative Adversarial Networks (CTGAN). Several regression models such as “Linear Regression”, “Decision Tree”, “Random Forest”, “Gradient Boosting”, “XGBoost” and “LightGBM” are trained and tested to find the most correct prediction model. The standard regression metrics including Coefficient of Determination (R²), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE) are used to assess the performance of the models. To increase transparency and trust in the predictions, SHAP (SHapley Additive exPlanations) is added to explain what role individual project features play in how much effort and development time they predict. The results of experiments have provided evidence that the ensemble learning models have better prediction accuracy and reliability than the traditional estimation approach. Proposed framework based on AI is an effective decision support tool for software project manager where they can get accurate effort estimates, optimal resource planning, decreased project risk, and optimized project scheduling. The findings of this study underscore the potential of combining machine learning, synthetic data generation, and explainable AI in the software project management and estimation landscape.