Next-Generation Human Activity Classification Using Hybrid Models for Digital Healthcare

Abstract

The need for accurate human activity recognition has increased with the rise of wearable devices and smart healthcare systems. This project focuses on developing a web-based application that can identify human activities using both machine learning and deep learning techniques. The system is designed in a simple way so that users can easily register, log in, and use different features without difficulty. After logging in, users can load the Kuhar dataset and view basic details such as total records, features, and activity labels. One important step in this work is feature selection, where Extreme Learning Machine (ELM) is used to reduce the large number of features into a smaller and more useful set. This helps in improving performance and reducing processing time. The system then applies different models along with the proposed ELM-GRU-AM model. From the results, it can be observed that the proposed model gives better accuracy compared to others, reaching around 96%. The system also shows performance using metrics like accuracy, precision, recall, and F1- score, along with confusion matrix graphs for better understanding. Users can also upload new test data and get predictions instantly. Overall, the system provides a simple and effective way to recognize human activities with good accuracy. Keywords—Human activity recognition, sensor data analysis, machine learning, deep learning, ELM-GRU-AM model, Extreme Learning Machine (ELM), feature selection, dimensionality reduction, GRU, attention mechanism, classification, Kuhar dataset, data preprocessing, training and testing, performance evaluation, accuracy, precision, recall, F1-score, confusion matrix, model comparison, real-time prediction, activity classification, web-based application, user interface