Testing Of Deep Learning-Based LSTM Model For Number Of Road Accidents Predicting

Abstract

Many have used the prediction of the number of road accidents, but it is still rare to find those who use and test prediction models that are not suitable. Predictive models that have been used to predict road accidents have proven successful, but have not provided model testing with data that is different from the deep learning approach. The LSTM model test is proposed to be tested with 5 different datasets from Kaggle and 3 hidden layer variations. The test results of the LSTM model are that with variations of 4 hidden layers it can achieve higher accuracy results than those without hidden layers and 2 hidden layers. The results are obtained from stability with the lowest average MSLE value and relatively balanced average time. Deep learning-based LSTM model testing was carried out to ensure and prove the stability of the model for predicting the number of road accidents in the future. Stakeholders can predict the number of road accidents using the resulting prediction model.