Human Activity Classification Using Supervised Machine Learning Algorithms

In this paper, using smartphones the human static and dynamic activities are measured via the inbuilt inertial measurement sensors. The dataset obtained from the experiments carried out on thirty individuals contains all the data regarding the angle change with different activities like walking, laying, sitting, etc. The machine learning techniques are applied to this dataset right from the onset of data cleaning to eliminate outliers and missing values and exploratory data analysis (EDA) to visualize the data and find relation between the variables. For eliminating the redundant data and improve execution time, a feature engineering technique called principal component analysis (PCA) is employed on the dataset. Finally, machine learning algorithms such as logistic regression, support vector machine and linear discriminant analysis are applied to the processed dataset. The accuracy is compared with all the algorithms with logistic regression outperforming the other two mentioned classifiers. These classifier models are found to be useful in applications where it is vital to monitor sensor data such as elderly monitoring, patient rehabilitation, fall detection and so on.