Development of a Hybrid Deep Learning Model for Car Crash Prediction Using Driver’s Behavioral Pattern
摘要
Road accidents is a major problem. It poses a huge threat to the driver, the passengers, the pedestrians, and the society in general. The World Health Organization (WHO) identified crashes involving vehicles as one of the most common causes of mortality in the world, causing considerable losses to our country’s current and future productive manpower, as well as major societal issues, fatalities, or catastrophic damage in many situations. The Car Crash Prediction System was created using three different algorithms to predict crashes before they occur so as to save lives and properties. The three algorithms used are the Long Short Term Memory (LSTM) algorithm, the Stacked Auto-Encoder (SAE) algorithm and a hybrid of both the SAE and LSTM (SAE-LSTM) algorithms. The data used was gotten from kaggle.com and vincentarelbundock.github.io. The dataset has a data size of 8,020 after cleaning. After training, the LSTM had an accuracy value of 98.6%, the SAE had an accuracy value of 76.4%, and the SAE-LSTM had an accuracy value of 81.4%. The result shows that LSTM outperform the SAE and the hybrid techniques. Future studies of car crashes and how drivers’ behavior affects it are needed to explore more variable that may cause accidents.