A Novel Deep Learning Model for Smartphone-Based Human Activity Recognition

Traditional pattern recognition methods have shown improvements in recent years. However, these techniques relied on human intervention to identify crucial information within the data. Deep learning has revolutionized this scenario by enabling computers to learn from data autonomously. This capability is precious for comprehending how individuals interact with mobile and wearable technology. The growing popularity of this method stems from its capacity to function effectively with minimal or no human involvement. This study introduces a novel hybrid deep learning network that merges Convolutional Neural Network (CNN) and Multi-Layer Perceptron (MLP) layers. While maintaining the ability to perform accurate activity identification, this CNN-MLP approach captures the temporal features from sensors, facilitating, therefore, multi-class classification. We also explore various machine learning models, such as Random Forests (RF), K-Nearest Neighbors (KNN), Logistic Regression (LR), Long Short-Term Memory (LSTM), and CNN-LSTM, on two well-established datasets: the WISDM and UCI HAR datasets. Through extensive testing on these datasets, we showcase the superior accuracy of our proposed CNN-MLP model compared to other competing machine learning and deep learning models. Our research opens up new possibilities for precise and effective human activity recognition on smartphones.