AI based Admission prediction model: Testing & Sustainability

This study aims to evaluate and compare the performance of various AI modelsRNN (Recurrent Neural Network), DBN (Deep Belief Network), DBM (Deep Boltzmann Machine), RBM (Restricted Boltzmann Machine), ESDRBM (Extended Stacked Denoising Restricted Boltzmann Machine)—using original and synthetic datasets to determine the viability of synthetic data for maintaining model performance. The objective is to assess whether synthetic data can serve as a reliable substitute for original data in training AI models, by comparing key performance metrics including precision, recall, FMeasure, accuracy, sensitivity, and specificity. The performance of each model was measured across the specified metrics using both original and synthetic datasets. A paired sample t-test was employed to statistically analyze the differences between the metrics obtained from original and synthetic data, assessing the significance of any observed differences. The results indicate minimal differences in performance metrics between original and synthetic data for all models. ESDRBM, RBM, and DBM consistently showed slightly higher precision, recall, and FMeasure values. DBM achieved the highest accuracy, while sensitivity and specificity metrics remained nearly identical across both data types. Paired sample t-tests confirmed that the differences between the original and synthetic data were not statistically significant, with high p-values indicating random variation as the likely cause of any observed differences. The findings suggest that synthetic data can effectively maintain the performance of AI models across various metrics. ESDRBM, RBM, and DBM models particularly exhibit robust performance with both data types. This underscores the potential of synthetic data as a viable alternative to original data in AI model training, providing flexibility and scalability in data generation without compromising model effectiveness.