Transportation mode detection using machine learning techniques on mobile phone sensor data.

The everyday use of means of transportation by millions of people combined with the continuous spreading of smartphones which are now equipped with various sensors, imply the existence of abundance of real-world transportation-related data and make Transportation Mode Detection (TMD) an interesting research field, essential to urban transportation planning, development of context-aware applications and physical and mental health improvement. The main objective of this work is to develop a machine learning methodology for classifying eight different transportation modes, including: still, walk, run, bike, car, bus, train, and subway, using data from smartphones sensors. To this end, publicly available datasets were used. For example, a subset of the original SHL dataset, including data obtained from one participant's smartphone embedded sensors (accelerometer, magnetometer, gyroscope, pressure sensor), being recorded for 68 days. As classifiers, eight Machine Learning algorithms were employed. The classifiers were firstly developed without Dimensionality Reduction (DR) and then with a DR feature extraction algorithm (Principal Component Analysis - PCA) so as to explore the possibility of using lighter models and potentially improve performance. After dimensionality reduction, the algorithms that performed best, accomplished a very good classification result in all classes while training time was significantly reduced.