Using Different Similarity Measures For Unsupervised Activity Recognition With Smartphone Accelerometers

The rapid popularity of smartphones is playing a more and more important role in people's daily life.It can be said with certainty that smartphones have been an indispensable part of our life.with the characteristics of powerful,having no additional cost and easy to be acceptd had led to a growing research interest in human activity recognition(HAR)with the mobile devices.Accelerometer is the most commonly used sensor of smartphone for HAR.Our experiment is designed according to the basic process of human activity recognition.We use self-collected daily behavior data sets and daily behavioral data sets exposed by UCI,which can make a comprehensive research on HAR.In feature extraction,we compare three different feature extraction methods,including frequency-domain feature extraction method,time-domain feature extraction method and mixed domain feature extraction method to find a more suitable feature extraction method for HAR based on smartphone.In activity recognition,many supervised HAR methods have been developed.However,it is very difficult to collect the annotated or labeled training data for HAR.So,developing of effective unsupervised methods for HAR is very necessary.The accuracy of an unsupervised method,such as clustering,can be greatly affected by the similarity or distance measures,because the learning process of clustering method is completely depending on the similarity between objects.In our experiment,Jaccard distance,Euclidean distance,Manhattan distance,cosine distance,Mahalanobis distance,Pearson correlation coefficients are applied to unsupervised behavioral recognition.In the experiments,the results of the different distance measures are compared,using three different feature extraction methods which include time-domain,frequency-domain and mixed-domain feature extractions.To comprehensively analyze the experimental results,two different evaluation methods are used:(a)C-Index before clustering,(b)FM-index after clustering.Experiments show that,almost for every combination of the feature extraction methods and the similarity meatures,the Jaccard distance measure is consistently better than the other distance measures for unsupervised HAR.