Forecasting Road Accidents Using Deep Learning Approach: Policies to Improve Road Safety

The development of smart cities holds immense significance in shaping a nation's urban fabric and effectively addressing urban challenges that profoundly impact the economy. Among these challenges, road accidents pose a significant obstacle to urban progress, affecting lives, supply chain efficiency, and socioeconomic well-being. To address this issue effectively, accurate forecasting of road accidents is crucial for policy formulation and enhancing safety measures. Time series forecasting of road accidents provides invaluable insights for devising strategies, enabling swift actions in the short term to reduce accident rates, and informing well-informed road design and safety management policies for the long term, including the implementation of flyovers, and the enhancement of road quality to withstand all weather conditions. Deep Learning's exceptional pattern recognition capabilities have made it a favored approach for accident forecasting. The study comprehensively evaluates deep learning models, such as RNN, LSTM, CNN+LSTM, GRU, Transformer, and MLP, using a ten-year dataset from the esteemed Smart Road Accident Database in Hubballi-Dharwad. The findings unequivocally underscore LSTM's superiority, exhibiting lower errors in both yearly (RMSE: 0.291, MAE: 0.271, MAPE: 6.674%) and monthly (RMSE: 0.186, MAE: 0.176, MAPE: 5.850%) variations. Based on these compelling findings, the study provides strategic recommendations to urban development authorities, emphasizing comprehensive policy frameworks encompassing short-term and long-term measures to reduce accident rates alongside meticulous safety measures and infrastructure planning. By leveraging insights from deep learning models, urban development authorities can adeptly shape the urban landscape, fostering safer environments and contributing to global safety and prosperity.