MACHINE LEARNING USING A COMBINED ARIMA MODEL AND LSTM FOR TRAFFIC DENSITY PREDICTION

• Main Author: Rahmi Maulida, Nabila
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/64379
• Institution: Institut Teknologi Bandung

Traffic regulation is often done to overcome congestion caused by overcrowding and roads that have excess capacity. However, this arrangement still utilizes information obtained from various entities on the road, namely the police and transportation service officers. Observation of the conditions and situations on the road is still subjective so that traffic management becomes subjective Smart city is a new breakthrough to help city problems, especially overcrowding. With these problems and opportunities, the density prediction system becomes a means of controlling density and becomes part of the Intelligent Transportation System (ITS). In this thesis research, a traffic density prediction model was built by combining the Autoregressive Integrated Moving Average (ARIMA) and Long Short Term Memory (LSTM) models. ARIMA was used in the first modeling to assist prediction in linear form. LSTM is used to complete the nonlinear prediction form according to the density characteristics of the road. This study uses the CRISP-DM methodology which is divided into five stages, namely understanding business needs, understanding data, cleaning and preparing data, optimizing parameters and modeling, and evaluating. In the data processing stage, the data is balanced by undersampling. In parameter optimization and modeling, a grid search with k-fold cross validation is used. In the evaluation stage, 2 metrics are used, namely RMSE and MAPE. This study was conducted with California traffic data. The results of this study indicate that the algorithm provides more optimal results with an RMSE value of 954 and a MAPE value of 101,013.