Author:
Andrea Vassilev
Affiliation:
CEA and LETI, France
Keyword(s):
Context Awareness, Transportation Mode, Data Mining, Classification, Smartphone, Sensors, Principal Component Analysis, Mahalanobis Distance, Linear Discriminant Analysis.
Abstract:
The recent increase in processing power and in the number of sensors present in today’s mobile devices leads to a renewed interest in context-aware applications. This paper focuses on a particular type of context, the transportation mode used by a person or freight, and adequate methods for automatically classifying transportation mode from smartphone embedded sensors. This classification problem is generally solved by a searching process which, given a set of design choices relative to sensors, feature selection, classifier family and hyper parameters, etc., find an optimal classifier. This process can be very time consuming, due to the number of design choices, the number of training phases needed for a cross validation step and the time necessary for one training phase. In this paper, we propose to simplify this problem by applying three data mining tools - Principal Component Analysis, Mahalanobis distance and Linear Discriminant Analysis - in order to clean the data, simplify th
e problem and finally speed up the searching process. We illustrate the different tools on the transportation mode classification problem.
(More)