LONG SHORT-TERM MEMORY RECURRENT NEURAL NETWORKS FOR SHORT-TERM TRAFFIC PREDICTION AT ROAD INTERSECTIONS

Abstract

The ability to predict short-term traffic designs enables Intelligent Transport Systems to proactively address potential events before they occur. Given the exponential growth in the volume, quality, and granularity of traffic data, novel techniques are necessary to effectively leverage this information to yield better outcomes while accommodating the ever-increasing data volumes and expanding urban areas. This study proposed a Long Short Term Memory (LSTM) Recurrent Neural Network for traffic prediction at road junctions was proposed and designed for short-term road traffic density prediction utilizing Long Short-Term Memory (LSTM) recurrent neural networks in implementation. The model was trained using the following dataset, vehicle ID, time of the day, vehicle type, weather condition, vehicle type and vehicle condition, obtained from road junctions and kaggle online dataset. The model was evaluated using the stated evaluation metrics, RMSE, SSE, R-Square, and R-Square Adjusted. The following results were obtained; RMSE was 0.128, SSE was 11.406357765197754, R-Square was 0.8670005614171354, and Adjusted R-Square was 0.8570256035234206.