Problem Statement : The client wanted to identify users from their distinct walking patterns. This was to be based on the raw motion data collected from mobile phones.Each person has a distinct walking pattern which is not easily recognised by the human eye. We had to make machines do it. There is an accelerometer and gyrometer in smartphones that capture the motion of the device. When a user keeps the phone in his pocket and walks, the device can capture the raw motion data of how the user is walking.
Our Solution : We have to do a lot of exploratory data analysis and pre processing before we could use the raw data. After that is done we engineer relevant features from the processed data that could be used to train a classifier.We solution from a gamut of latest classifiers** and their different variants to reach the most optimised solution.
This is how the raw data from 2 different users looks like. It is difficult even for humans to distinguish but with ultra advanced ML feature engineering, time series techniques and latest AI algorithms, patterns can be identified and these patterns help in in distinction between users.