This study presents a robust framework for individual pharmacy cost prediction using scalable hybrid deep learning models. Utilizing the 2017 Medicaid State Drug Utilization dataset, the research addresses challenges such as data sparsity, heterogeneity, and high dimensionality. The proposed models include traditional algorithms—Random Forest (RF), Logistic Regression (LR), Support Vector Machine (SVM), Artificial Neural Networks (ANN)—and advanced deep learning techniques like LSTM, Bi-LSTM, GRU, Autoencoder, and a novel hybrid Auto-RNN (Autoencoder + RNN). A TabTransformer model is also evaluated for tabular data handling. Among them, two hybrid models—Auto-GRU)—demonstrate superior predictive performance in terms of MSE, MAE, and RMSE, both with and without feature correlation. The study also integrates LIME for explainable AI, enhancing model interpretability. This hybrid modeling approach significantly improves scalability, accuracy, and explainability, offering a valuable tool for healthcare providers and policymakers to optimize pharmacy-related expenditures.

Keywords:
Pharmacy Cost Prediction, Hybrid Deep Learning, Auto-GRU, Medicaid, Autoencoder, GRU, LSTM, Bi-LSTM, TabTransformer, Explainable AI, Healthcare Analytics, Cost Forecasting.