LPBSA: Pre-clinical data analysis using advanced machine learning models for disease prediction

Abstract

Diabetes, COVID-19, and heart disease pose significant global health challenges. The current study introduces an optimization algorithm, Learner Performance-Based Behavior with Simulated Annealing (LPBSA), integrated with Multilayer Perceptron (MLP) as a neural network technique to improve disease prediction accuracy. The algorithm was tested on six preclinical datasets (one is related to diabetes, two are related to heart disease, and three are related to COVID-19). In addition to LPBSA-MLP, other optimization algorithms, including Fitness Dependent Optimizer (FDO), the original Learner Performance-Based Behavior (LPB), were independently combined with MLP. Furthermore, all three algorithms were integrated with a Cascading Multilayer Perceptron (LPBSA-CMLP, FDO-CMLP, LPB-CMLP) to enhance the convergence speed and learning capability. This allowed for a comprehensive comparison across diverse algorithmic configurations and enabled the identification of the most efficient model for disease prediction. The proposed LPBSA-MLP model achieved 100% accuracy on four data sets and at least 99.31% on the others. Further metrics confirm its performance: sensitivity and specificity values reached 100%, and Mean Square Error (MSE) values ranged from 0.0008 to 0.003. When benchmarked against models trained with FDO-MLP, LPB-MLP, and other standard optimizers, LPBSA-MLP consistently outperformed them in terms of both classification performance and convergence speed. These findings indicate the effectiveness of LPBSA in enhancing predictive modeling for critical health conditions.