Recurrent Neural Networks And Crisp-Dm In Traffic Accident Analysis And Prediction In Bucaramanga, Colombia.

Abstract

The work aimed to implement recurrent neural networks (RNN) using the CRISP-DM methodology to analyze and predict the severity, monthly frequency, and daily number of people involved in traffic accidents in Bucaramanga, Colombia. Data collected from January 2012 to September 2023 were used. The CRISP-DM methodology guided all phases of the project, from business and data understanding to modeling and evaluation. RNN models, including Many-to-One and LSTM (Long Short-Term Memory), demonstrated high accuracy in classifying accident severity and moderate accuracy in regressing the number of accidents and people involved. The results provided a valuable tool for traffic authorities in improving road safety. These findings highlight the utility of RNNs and the CRISP-DM methodology in analyzing and predicting traffic accidents, offering a solid basis for informed decisions in road safety management.