IoT-Based System for Human Localization Activity Recognition Using Hybrid Deep Learning Techniques

Abstract

Indoor positioning and navigation is an emerging field where accurate location identification and activity recognition with precision are important factors. Due to the emergence of handheld devices with location enabled and their importance in smart homes, industries, health monitoring, and security surveillance, human activity localization is also of great concern and importance. Keeping in mind the importance of accuracy with precision, we have proposed an IoT-based solution for human activity recognition using a hybrid deep learning approach in this article. In our proposed model, we have integrated Convolutional Neural Networks (CNNs) with advanced feature extraction, with an added feature of optimization for enhanced accuracy and precision. Our proposed hybrid model successfully classifies human activities such as “AT SCHOOL,” “LOC Home,” “Indoor,” and “Outdoor” using IoT-based sensor data received from multiple stations. The accuracy of our proposed deep learning hybrid model is 96%, compared to existing techniques for human activity recognition such as Deep Neural Decision Forest (89%), HAR-graph CNN (87%), and Random Forest (87%), with enhanced precision, recall, and F1 score, respectively. For data augmentation and optimization, we have used SMOTE and Yeo-Johnson to address the issues of class imbalance and feature distribution, respectively. Moreover, 5-fold cross-validation is used to ensure the robustness and efficiency of the proposed model for localizing human activities with enhanced accuracy and recognition.