LSTM model for high-risk COVID-19 transmission among Thailand’s mass rapid transit purple line passengers

Abstract

Modern epidemiological research increasingly integrates machine learning and data-driven methods to enhance the prediction and mitigation of COVID-19 and other respiratory virus outbreaks. Herein, a long short-term memory (LSTM)-based classification model was developed to predict high-risk COVID-19 transmission zones on Thailand’s Mass Rapid Transit Purple Line platforms. Using sequential passenger flow data and temporal patterns, platform areas were classified into low- and high-risk areas based on key inputs including station, date, time, and crowd density. Hyperparameter optimization using RandomizedSearchCV yielded an optimal configuration of 64 LSTM units, a learning rate of 0.001, a batch size of 32, and 30 epochs. The model achieved 98% test accuracy, 98.22% cross-validation accuracy, and 99.11% peak validation accuracy. For high-risk detection, it obtained precision and recall of 0.95 and 0.96, respectively. The results highlight the robustness and real-time applicability of the approach in urban transit systems. The findings offer actionable insights for targeted interventions such as dynamic crowd management and optimized resource allocation, thereby reducing exposure risks and strengthening preparedness for future public health crises.