Abstract
Software cost estimation (SCE) is among the most critical task in software development and project management, that can directly impact the budgeting, planning, and resource allocations. The current study proposes a hybrid model combining TabNet, a deep learning architecture designed for tabular data, with Harris Hawks Optimization (HHO) for feature engineering. HHO algorithm is being used to identify the most significant feature among all the input and construct informative feature combinations that enhance model performance. Then the TabNet is used in analyzing the dataset for predicting the cost incurred in software development process. The TabNet-HHO framework is evaluated on standard datasets, including COCOMO and NASA project data. Explainable AI (XAI) technology based on SHAP is used in the current study to analyze the feature contribution in the decision process. Furthermore, the dependencies graphs are presented to analyze the relationships among the features. The proposed model is being evaluated using the standard metrics like mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), median magnitude relative error (MdMRE), and prediction accuracy (PA). The experimental outcome has proven that the proposed TabNet-HHO model has outperformed various existing software cost estimation models. The performance of the TabNet-HHO model is further evaluated across divergent datasets like Desharnais, China, Albrecht for SCE. TabNet-HHO has performed well across all the datsets, with a better prediction accuracy of 98.82% on evaluation over the COCOMO dataset.