This project aims to analyze a sizable dataset focused on Human Activity Recognition (HAR). The dataset includes sensing signals from accelerometers and gyroscopes to identify various human activities such as walking, standing, and sitting, turning it into a multi-class sequence classification problem. HAR is crucial for understanding daily activities and plays a significant role in health management. The dataset, available on MyAberdeen, originates from a research project exploring anomaly detection and events in pressurized water reactors.
- Classes: Six classes, including walking, walking upstairs, walking downstairs, sitting, standing, and laying.
- Features: Utilizes 3-axial linear acceleration (three features) and 3-axial angular velocity (three features).
The primary objective is to develop a set of classification models capable of automatically classifying human activities based on sequential sensing data (features). This project addresses a multi-class sequence classification problem without assuming prior domain knowledge.
-
3-axial Linear Acceleration:
- Feature 1
- Feature 2
- Feature 3
-
3-axial Angular Velocity:
- Feature 4
- Feature 5
- Feature 6
Explore the dataset and understand the feature representations. Implement classification models for human activity recognition. Evaluate and fine-tune models based on performance.
- Data Exploration: Understand the structure and content of the dataset.
- Model Development: Implement classification models.
- Evaluation: Assess model performance using appropriate metrics.
- Fine-tuning: Optimize models for better accuracy.