Smartphone transportation mode recognition using a hierarchical machine learning classifier

The paper develops a hierarchal classification framework for transportation mode recognition. The 9 proposed approach uses different supervised learning methods from the field of machine learning as 10 building blocks. This framework is used to distinguish between different transportation modes including 11 driving a car, riding a bicycle, taking a bus, walking, and running. The proposed framework consists of 12 two layers. The first layer consists of only one multiclass classifier while the second layer consists of a 13 pool of binary classifiers. Each classifier uses a subset of features that produces the best discrimination 14 power between target classes. Each subset is chosen using the minimum Redundancy Maximum 15 Relevance (mRMR) method. The proposed approach is trained and tested using data obtained from 16 smartphone sensors including accelerometer, gyroscope, and rotation vector sensors. The experimental 17 results show a promising performance of the hierarchical framework when compared to the traditional 18 classifiers. Furthermore, the hierarchical framework that uses the random forest (RF) classifier in the first 19 layer and Support Vector Machine (SVM) classifiers in the second layer yields a 96.32% classification 20 accuracy, which is better than the recent reported results