Learning transportation modes from raw GPS data

In a computing environment, a system comprising:

• a segmentation mechanism configured to separate GPS data into a plurality of segments;
• a feature extraction mechanism configured to extract features for respective segments;
• and
• an inference model configured to use at least some of the features extracted for respective segments to infer a transportation mode for respective segments.

One or more computer-readable media having computer-executable instructions, which when executed perform actions, comprising:

• processing GPS data into walk segments and non-walk segments based on velocity data or acceleration data, or both velocity data and acceleration data;
• determining a length of respective segments;
• comparing the lengths of the segments against a merge threshold distance, and merging any segment having a length below the threshold distance into another segment;
• comparing the lengths of the segments against a certainty threshold distance;
• and for any uncertain segments that do not meet the certainty threshold distance, determining whether a number of consecutive uncertain segments is present, and if so, merging the consecutive uncertain segments into a non-walk segment;
• and
• determining a predicted mode of transportation for respective walk segments and for respective non-walk segments, based on features of respective segments.