Predicting Traffic Status of Inner Bangkok with Deep Learning

Abstract

The objectives of this research was to compare the efficiency of predicting road traffic status in the inner Bangkok area, between the properly parameterized deep learning models with different conventional algorithms. In process of research methodology, the researcher used a computer program developed to collect traffic situation images from Google Maps specific inner Bangkok area, every half hour for 2 months, 1,844 traffic images were obtained, combined with duration of the day to create more independent variables. The findings revealed that percentage of prediction accuracy when predicted with test data sets in different algorithms as follows, Decision Tree: 71%, Random Forest: 71%, K Nearest Neighbors: 73%, Naive Bayes: 63 %, Gradient Boosting: 72% and Deep Learning: 75%. This research proposed a practical method for collecting traffic data from images on Google Maps websites. Preprocessing data ready for machine learning and a model for predicting traffic status with deep learning. Which the result of the prediction is the road traffic condition predicted 1 hour in advance at the study point was the area of ​​Sathorn Road. Specify a prediction as a class, according to the color scale of Google Maps, there are 4 classes, namely, green means active traffic, orange means moving, red means heavy traffic and dark red means traffic is congested very much. The model predicts the most probable traffic status. The results of the predictions are useful for people traveling or transportation logistics companies who use routes to know the traffic status in advance, including those interested in studying traffic data with machine learning.