A Backpropagation-Based Artificial Neural Network Model for Predicting Pharmaceutical Demand

Drug inventory management is a vital component of the healthcare system because it ensures the continuity of essential drug supply and pharmaceutical logistics efficiency. However, most pharmaceutical facilities still rely on manual forecasting methods based on historical trends that are linear in nature and unable to capture the nonlinear relationship between morbidity rates and drug demand. As a result, there is a mismatch between stock and actual demand, leading to shortages or surpluses and an increased risk of drug expiration. This study aims to develop an artificial neural network (ANN) model for predicting drug demand using a backpropagation algorithm to improve the accuracy of estimates and the effectiveness of stock planning. The data used included five years of drug usage records and the prevalence of the ten most common diseases in the Pharmacy Installation. The model was designed with a multilayer perceptron architecture (25–70–25–1) using a log-sigmoid activation function and a trainCGF training algorithm. The training results showed optimal performance with 94.2% accuracy, MSE 0.0135873, and MAPE 5.793%, accompanied by a strong correlation between the target and output (R = 0.99935). This demonstrates the model's ability to learn nonlinear patterns and produce stable and reliable predictions. The implementation of the JST model enables the optimization of drug distribution by reducing the risk of stockouts and overstocking, while also reducing waste due to expiration. This prediction system has the potential to become an adaptive and sustainable decision-making tool in public pharmaceutical supply chain management, in line with the principles of resource efficiency and sustainability of health services.