Exploration and Deduction of Sensor-Based Human Activity Recognition System of Smart-Phone Data

Human activity data measurements are identified by using 3 axial accelerometer and gyroscope raw signals from wearable inertial sensor data. The smartphones are used to recognize the human activities. Numerous android applications utilizing accelerometer are accessible on a smartphone which has wide deceivability and potential commercial centre. This proposed work, focuses on only one inertial sensor data, that the smartphone is attached to the waist sensor. This sensor data proves to provide good results among other sensors. This work handles UCI HAR Smart Phone dataset has data with 30 volunteers at age difference of 19-48 years, and with 561 measurements variable. In this paper Data Mining based techniques are proposed to exaggerate the performance measures of the human activity recognition system using data preprocessing, feature selection, feature extraction, five different types of classification: Decision tree, KNN, Naive Bayes, Support Vector Machine and Random Forest, used to recognize the human activities. The Random Forest classification method emerges and classifies with 100 % accuracy and also predicts the movement of the human which is isolated into six categories of activities such as sitting, standing, laying, walking, walking-upstairs and walking-downstairs. The performance evaluation is analyzed by comparing the F-Score, accuracy and kappa values of various classification models.