Real Time Map Matching for Low Frequency GPS Data Based on Machine Learning Technology

GPS cannot perform well in urban area because of multipath problems that make GPS location contain low accuracy. So, the map matching method is the method to adjust GPS location to reduce GPS location error. However, the current map matching method requires GPS high frequency data for adjustment and it does not perform well in GPS low frequency data. So, the study explores the idea to improve the map matching method based on a kind of machine learning technology by using the classification method to adjust the low frequency GPS location. This research uses features such as speed, heading and location of the previous GPS point for a training machine learning the model and adjust points of the GPS location. The objective of this study is to develop accuracy of the map matching method and the speed of map matching method for low frequency GPS data. After processing data, we found that Random Forest is the machine learning model that contains highest precision, recall and F1-Score. Moreover, this model is a low sensitivity model when frequency of data changes, but Decision Tree is the fastest model in prediction time and it takes lower time for training than Random Forest model.