Neuro-Genetic System: A Hybrid System of CNN-BiLSTM Optimized by Genetic Algorithm for Road Accident Severity Prediction

The topic of road accident severity has received considerable focus in recent years. Despite advancements in traffic safety management, the prevalence of road traffic casualties continues to be a cause for concern. The increasing popularity of deep learning methods has led to their application in understanding this phenomenon. However, effectively tuning these models to obtain satisfactory outcomes poses significant challenges. This study presents a hybrid system that combines a Convolution Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) to predict road accident outcomes. The system incorporates a Genetic Algorithm for feature selection and optimizing hyperparameters. The model's results were compared to baseline learning classifiers. This study utilizes data on road accidents in the United Kingdom from 2000 to 2018. The study found that using a Genetic algorithm for feature selection and optimizing hyperparameters improved the performance of the model.