ENHANCING AI-DRIVEN SOFTWARE SCALABILITY WITH BAYESIAN NEURAL NETWORKS AND PARTICLE SWARM OPTIMIZATION FOR ROBUST DECISION-MAKING

Software defect prediction is critical for software reliability and quality. The traditional approaches, such as rulebased techniques and statistical models, have pitfalls like low generalization ability, high false positive rates, etc.,in addition to their inherent incapacity to deal with very complex, high-dimensional data. Machine learningapproaches, for example, Decision Trees and Support Vector Machines (SVMs), are being researched, yet theytoo have antagonistic aspects like hyperparameter optimization and less optimal accuracy and adaptability. Thecurrent work presents a Bayesian Neural Network (BNN) optimized using Particle Swarm Optimization (PSO) toovercome these hurdles for software defect prediction. The proposed framework involves collecting data, doingpre-processing (data cleaning and normalization), model training via BNN, and hyperparameter optimization withPSO, which fine-tunes the parameters of the model efficiently to ensure improved accuracy and generalization.Experiments show that the BNN + PSO model has a much higher performance compared with traditional methodswith 99.5% Resource Allocation Efficiency, 99.2% Data Processing Speed, 99.8% Optimization Accuracy, 98.9%Adaptability, and 98.7% Error Reduction. The findings reveal that the model shows superior performance inclassification, lower false acceptance rates, and efficient resource utilization, which makes it an effective modelfor real-time defect prediction in vast scales of software development environments. The present study combinesBNN with PSO to provide a very robust, scalable, and efficient solution to the issues of software defect prediction,thereby providing significant advantages over the current status methods