Performance Analysis of Machine Learning Algorithms for Smartphone-Based Human Activity Recognition

As the number of smartphone users is increasing exponentially, there is an increase in the availability of continuous sensor data, which has attracted enormous interest in sensor-based human-activity recognition (HAR). Recognizing human activities is particularly important in detecting abnormal activities and tracking a person's physical activity, especially in healthcare applications, among many others. In this paper, HAR analysis is conducted with three different machine learning algorithms (Support Vector Machines (SVM), Decision tree, and random forest methods) based on smartphone sensors. Machine learning algorithms are capable of identifying and differentiating between different human activities using mobile phone sensor data. The smartphone sensors (gyroscope and accelerometer) data are recorded at the Koneru Lakshmaiah Education Foundation University campus, Guntur, India, with different human activities. In this research work, the data from smartphone mobile sensors were initially analysed with SVM, decision tree, and random forest algorithms. To evaluate the machine learning algorithm's accuracy, F1 score for different smartphone sensors for both individually and combined is estimated. The results indicate that the proposed machine learning methods can derive a relation between type of activity, algorithm, smartphone sensors data, and their corresponding accuracy. The outcome of this work would be beneficial for detecting abnormal features of older people with a smartphone device.