Human Activity Recognitions in Handheld Devices Using Random Forest Algorithm

Human activity recognition (HAR) using embedded sensors are a rapidly developing topic with several uses, such as fitness tracking, health monitoring, and context-aware services. The data is categorized into known activities, including sitting, standing, lying, walking, and walking upstairs and downstairs. The accelerometer and gyroscope of the device were utilized to produce sensor data, and noise filters were employed to pre-process the sensor signals. Using a Butterworth low-pass channel, the sensor speed increase signal—which consists of both body movement and gravitational components—was separated into two components: body speed increase and gravity. We expect just low-recurrence components in the gravitational power. Through the computation of factors from time and recurrence space, a vector of highlights was obtained. Based on smart phone sensor data, this paper investigates the application of machine learning methods to identify and classify human activities. A crucial part of the examination includes the choice of proper AI calculations. During the model training phase, the preprocessed dataset is presented to the chosen algorithm, allowing it to learn intricate patterns and relationships between sensor data and human activities. A far-reaching assessment of model execution follows, utilizing standard measurements like exactness, accuracy, review, and F1 score. Hyperparameter tuning refines the model for ideal outcomes.