Topological Feature Extraction For Human Activity Recognition Using Persistent Homology

Abstract

The focus of this article is to investigate the effectiveness of Topological Data Analysis (TDA) for Human Activity Recognition (HAR) using smartphone sensor data. Using persistent homology, topological characteristics were extracted from the Human Activity Recognition Using Smartphones dataset. Representative persistence intervals were chosen using the AvgInt sampling technique. Three machine learning methods were then used to classify the retrieved topological descriptors: Random Forest (RF), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM). Confusion matrix-based classification accuracy and scatter plot visualizations were used to assess the suggested framework. According to experimental findings, TDA-derived features are highly discriminative for classification and efficiently capture activity-related patterns. Among the classifiers that were assessed, Random Forest demonstrated relatively lower accuracy, whereas SVM and KNN reached the same classification performance. These findings demonstrate the potential of integrating TDA with machine learning techniques for smartphone-based activity recognition and highlight the importance of classifier selection when working with topological feature representations.