A New Post Correction Algorithm (PoCoA) for Improved Transportation Mode Recognition Zelun Zhang School of Electronic Engineering and Computer Science Queen Mary, University of London London, UK zelun.zhang@eecs.qmul.ac.uk Stefan Poslad School of Electronic Engineering and Computer Science Queen Mary, University of London London, UK stefan@eecs.qmul.ac.uk Abstract—Transportation mode plays an important role in enabling us to derive a mobile user’s context, and to adapt intelligent services to this. However, current methods have two key limitations: a low recognition accuracy and coarse-grained recognition capability. In this paper, we propose a new Post Correction Algorithm (PoCoA) that is applied after the use of typical classifiers to address these limitations. We evaluated the use of PoCoA for the following transportation modes, walking, cycling, bus passenger, metro passenger, car passenger, and car driver. PoCoA enhances a typical accelerometer-based transportation recognition method with a more accurate sub- classification of motorized transportation modes when tested on a dataset obtained from 15 individuals. Overall accuracy improved from 69% to 88% when comparing with a state of the art two- stage classifier (Decision Tree + Discrete Hidden Markov Model). Keywords- Transportation mode recognition; Accelerometer; Post Correction Algorithm; Hidden Markov Model (HMM) I. INTRODUCTION Transportation mode is an important type of user context that denotes someone’s mobility status while travelling. In this research, we are interested in recognizing the urban transportation modes of normal adults. The transportation modes include walking, cycling, bus passenger, tube (London‘s Metro or Underground Train) passenger, car/taxi passenger, and car driver. Such transportation mode recognition could facilitate a range of applications as follows. Human-Centered Activity Monitoring: Transportation modes of individuals can be logged and mapped to locations to enable individuals to plan travel based on physical activity targets and use in health monitoring [1], e.g., a mobile phone may detect how many hours a person walks every day and to provide personalised health advice [2]. Individual Environmental Impact Monitoring: The transportation mode can be used to provide a personalized environmental scorecard for tracking the environmental impact of one’s activities. Examples include Personal Environment Impact Report (PEIR) and UbiGreen [3, 4], along with commercial offerings such as Ecorio and Carbon Diem [5, 6]. Distributed Intelligent Services Adaptation: In-situ information can be adapted to mobility profiles (time, location, transportation mode traces), which can contribute to distributed user-context and group context-awareness [7], e.g., to automatically adapt a navigation map from a bus-route view while someone is on a bus, to a pedestrian view triggered by starting to walk from the destination bus stop to a meeting place. Implicit Human Computer Interaction: Transportation mode recognition in people’s daily life can help to enable the hidden computer part of the vision of intelligent transportation system in terms of reducing a user’s cognitive load when interacting with services while travelling, e.g., a mobile phone may detect when a person is driving or involved in vigorous physical activity, and automatically divert a call for safety considerations [8]. The confluence of advanced wearable sensor technology embedded in widely available mobile phones offers the opportunity for automatic recognition of a person’s activities and transportation modes in daily life [9]. Mobile phone integrated accelerometer and GPS can provide spatial user contexts (e.g. acceleration and speed) during different activities in real time [10]. Typically, thresholds for speed and acceleration are often used to differentiate transportation modes [8]. However, under certain traffic conditions, i.e., congestion, speed and acceleration for different transportation modes can coalesce making the modes hard to be differentiated using such thresholds alone, e.g., to sub-differentiate motorised modes with a high accuracy. Hence, additional post-acquisition analysis is needed to supplement threshold-based analysis, to possibly reclassify transportation modes (two-stage classifier). The main contributions of this paper are: First, we reproduced the results of a best practice method (an accelerometer-based method identified in the survey) to be used as a baseline method to evaluate our method through observing 15 different individuals’ daily mobility patterns. Second, we proposed a novel Post Correction Algorithm that is applied after the use of typical classifiers. We then evaluated this experimentally through comparing PoCoA with a more typical two-stage classifier (DT+HDMM). Third, after generating more comprehensive datasets by simulating the classification process of the accelerometer-based method, we further demonstrated the potential usefulness of our new Post Correction Algorithm. II. RELATED WORK Recognising transportation mode through sensing modalities that are available on mobile phones, mainly the accelerometer and GPS, has been the subject of much research. 978-1-4799-0652-9/13/$31.00 ©2013 IEEE