Cost-Intel: A Hybrid Deep Learning Framework using TabNet, HHO, and SHAP for Explainable Software Project Management

The essential issue of software cost estimation is a very important one for assessing the viability of a project, however, modern software cost estimation models often have difficulties with large data sets and lack algorithmic transparency. In this research, a Hybrid computational framework is proposed that combines Cost-Intel, a new hybrid framework of Dynamic Feature Space reduction based on the Harris Hawks Optimization (HHO) with TabNet, an interpretable deep tabular learning architecture. The system is able to identify high value cost drivers, and results in a Mean Magnitude of Relative Error (MMRE) of 0.1066, which is far more accurate than conventional algorithmic models. Beyond this, the framework incorporates a Retrieval-Augmented Generation (RAG) pipeline, which will ingest real GitHub source code, automatically generating Work Breakdown Structures, thus seamlessly bridging predictive AI with actionable project management.