Traffic Incident Detection in Jakarta on Twitter Texts Using a Multi-Label Classification Approach

Congestion is a phenomenon or traffic incident that often occurs in big cities. These incidents certainly can be detrimental to existing road users. Therefore, it is necessary to detect traffic incidents. Twitter is one of the media that can be used to find information about an incident in a real-time. In this study, a multi-label classification was carried out using data sourced from the Polda Metro Jaya TMC Twitter account to identify various traffic situations, such as smooth traffic, heavy/jammed traffic, weather conditions, and traffic accidents. This study uses machine learning with Support Vector Machine, Complement Naive Bayes and Logistic Regression, deep learning with Long-short Term Memory, and transfer learning from a pre-trained model IndoBERT. Based on the evaluation results of classification models, IndoBERT has the highest evaluation score with 99.10% f1-score and 99.26% accuracy.