Predictive Analysis of Daily Activities on A Mobile Device using Shallow CNN

An emerging area of artificial intelligence and machine learning called "Human Activity Recognition" (HAR) aims to create systems that can comprehend and categorize human movements and behaviors. HAR uses data captured from wearable devices, smartphones, and other sensors to infer activities performed by individuals. Leveraging sophisticated machine learning algorithms, HAR systems are capable of accurately identifying a wide variety of actions, from straightforward gestures to challenging physical demands. But a major obstacle to HAR’s current state is the lack of detailed data on how much time is spent sitting, standing, lying down, and being active. Existing wearable devices and health trackers primarily focus on providing inactivity alerts, leaving users with incomplete insights into their overall activity patterns and posture distribution throughout the day. This information gap hinders users’ ability to monitor and improve their sedentary behaviors, leading to potential health risks and suboptimal activity management. Reciprocating this challenge, the proposed solution aims to utilize motion sensor data from gyroscopes and accelerometers found in smartphones to determine position of a human using a Shallow Convolution Neural Network (SCNN) model.