Transportation mode detection based on mobile sensor data

This thesis addresses transportation mode detection based primarily on mobile phone data using machine learning methods. Our approach uses short samples of accelerometer readings taken while traveling in a vehicle to distinguish between three modalities --- car, bus, and train. We use gravity estimation to pre-process the samples. We extract features from statistical, frequency-based, and peak-based domain. With statistical analysis of the features we gain an introspective into the data. To additionally analyze the features we construct several feature sets for classification. As a classifier we use random forest, support vector machine, and neural network. Our approach correctly classifies 65% cars, 63% buses, and 18% trains using neural network