Human Understanding from Lifelog Data via Mobile Sensors Using a GRU Network with Attention Mechanism

Lifelog data, collected from smartphones and wearable devices, encompasses various types of information crucial for accurately assessing an individual's health status. Emotion analysis and prediction play a vital role in the early detection of mental health issues, while stress index prediction aids in preventing health problems related to chronic stress. This study proposes the Smart Sensor Prediction (SSP) model, which predicts an individual's emotional state, stress index, and sleep condition based on lifelog data collected using IoT technology. Our SSP model integrates Gated Recurrent Units (GRUs) and Attention mechanisms to capture the temporal correlations of sensor data with high precision. Through extensive data preprocessing and feature engineering, we extract meaningful information from diverse sensor inputs. The model demonstrates superior performance compared to existing methodologies, achieving a weighted F1-score of 5.96 in our experiments using real lifelog data. This research enhances model prediction accuracy by integrating various biosignals and environmental data.