Machine Learning in Motion: Enhancing Human Activity Recognition with Smartphone Sensor Data and Performance Metrics
S.Sreenath Kashyap, Gagandeep Gagandeep · 2024
The goal of this undertaking course, is to enhance the efficiency and outcome performance in Human Activity Recognition (HAR) using smartphone sensor data and machine learning (ML) approaches. This field of study is located at the interface between pioneering computational techniques and practical technical implementations. The rising demand for high-performance, robust and user friendly HAR systems is the driver behind this initiative. The importance of such systems is increasing not only in the fields of medicine, training, and cities development but also aided by these areas. This investigation will have a primary goal for appraising many ML models. The models include different types of neural network architectures, including support vector machine (SVM), decision trees (DT), convolutional neural networks (CNN) and long short-term memory (LSTM). In training and validation procedures, the models were given datasets based on information from sensors in smartphones accelerometers and gyroscopes. Precisely, the goal was to provide a precise identification and value of wide range human behaviors. One of the key components that make this course meaningful is its clear-cut inclination towards practical application in real world situation. To do this, giving priority to the accuracy of models is paramount while simultaneously taking into consideration parameters such as speed of processing and practical aspects when using most smartphones. For HAR tasks, CNN and LSTM achieve better results as opposed to other deep learning approaches. They have amazing ability to harness more or less characteristics of the vibrations, which naturally happen in sensor data. The findings from the studies show that deep learning approaches perform better than traditional ML models.