A Hierarchical Classifier for Detecting Metro-Journey Activities in Data Sampled at Low Frequency

Abstract

This paper aims to build a novel classification model that can distinguish the typical motion activities that a traveler would perform in a metro-journey. Following motion activities are focused in this work: waiting in a queue, traveling in a metro train, climbing-up, climbing-down, walking and stationary. We aim to build a classifier which can work on data sampled from smartphone sensors at a low frequency (4Hz). However, it is non-trivial to do so as the mentioned activities are not easily separable in data sampled at low frequency. Current works focus on data sampled at high frequency (40Hz). Also, they don't consider metro-journey specific activities such as queue. Our proposed model focuses on all the metro-journey specific activities while using data sampled at low frequency (4Hz). Experimental evaluation (datasets collected in Delhi Metro-rail network) indicate superior performance of our classifier (mean accuracy 92%) over the related work (mean accuracy 70%).