Performance Analysis of Machine Learning Models for Human Activity Classification

Abstract

One of the ultimate goals of machine learning is to design an application that can accurately understand 'humans' actions and intentions to serve us better. Hence, this research paper aims to explore the application of machine learning algorithms in human activity classification. It focuses on evaluating the performance of some popular machine learning methods, namely K-NN, Support Vector Machine (SVM), Logistic Regression and XGBoost algorithm, for classifying human activities. Hyperparameters of algorithms are tuned to find the optimized result and performance. The performance of K-NN is evaluated over a range of K, while the logistic regression is implemented with various solvers and penalty functions. Further, SVM is tested for multiple linear and non-linear kernel functions while the XGBoost algorithm is implemented with different learning rates and trees. On performance comparison, it is observed that all four Machine Learning Models have encountered ambiguity in identifying similar activities. Further, when implemented with a polynomial kernel, the SVM method outperformed other state-of-the-art techniques, achieving an accuracy of 96.9% and obtaining peak values of precision and F1-score. At the same time, Logistic Regression failed to classify several activities, leading to the lowest accuracy.