Clustering Analysis of Human Behavior Based on Mobile Phone Sensor Data

Abstract

In recent years, some of the methods based on the deep learning theory to identify human activities have been proposed. However, most of these methods are not suitable for hardware with limited resources when it comes to rapid real-time analysis considering its analysis complexity and data scale. In order to minimize the amount of input data, a variety of clustering analysis methods are proposed to address this issue. Six clustering algorithms including K-means clustering algorithm, K-mode clustering algorithm, CLARANS clustering algorithm, BIRCH clustering algorithm and DBSCAN clustering algorithm are implemented to reduce the feature scale, results in greatly reduced amount of input data while a high recognition accuracy is still ensured. When the DBSCAN clustering algorithm is used to reduce the feature dimension from 561 to 145, the classification accuracy drops merely less than 2%, reaching 93.58%. Even when the number of features is reduced to 100, the accuracy can still be maintained at 91.38%. By using Birch clustering algorithm, CLARANS clustering algorithm and K-mode clustering algorithm, the classification accuracy is 89.75%, 85.4% and 76.28% respectively.