Deep Learning — Based Feature Fusion and Estimation for Ensuring Software Resuability

The document addresses the challenges and objectives in the field of software reusability estimation, highlighting limitations of existing techniques and proposing a novel approach using a Deep Neuro-Fuzzy Network (DNFN) optimized by a Competitive Tunicate Swarm Optimization (CTSO) algorithm. Key challenges identified include insufficient static coupling measures for object-oriented metrics, difficulties in predicting software aging and cost for mobile applications, and the dynamic nature of decision-making processes affecting reusability. Additionally, current methods like integrated random forest and gradient boosting techniques lack meta-heuristic approaches and deep learning integration for better accuracy. The proposed methodology involves extracting significant reusability metrics, transforming data using the box-cox model, and performing feature fusion with Levenshtein distance and Deep Residual Network. The CTSO algorithm, integrating Competitive Multi-Verse Optimizer and Tunicate Swarm Algorithm, is employed to train the DNFN. Through a comparative analysis, the proposed CTSO-based DNFN demonstrates superior performance in terms of MAE, MMRE, and SEM when compared to traditional neural networks and fuzzy inference systems, indicating more accurate and reliable software reusability estimation.