This project develops a Wearable Sensor Data Collection Pipeline tailored for Spatiotemporal Adversarial Machine Learning within Human Activity Recognition (HAR). The project's focus on adversarial vulnerabilities in HAR systems, including evasion, poisoning, model inversion, and membership inference attacks, positions it at the forefront of addressing security challenges in machine learning. The project establishes a robust data processing pipeline using Python to handle a variety of sensor types including accelerometers, gyroscopes, magnetic sensors, and GPS. This setup includes functionalities for file parsing, noise cleaning using a Butterworth lowpass filter, statistical feature computation, and data aggregation, all enhancing the integrity and quality of data for training machine learning models. The design emphasizes modularity and scalability, facilitating the integration of additional preprocessing steps and feature extraction methods as needed.

The pipeline significantly improves data quality through advanced noise reduction techniques, which are crucial for developing reliable HAR systems. It supports a comprehensive suite of features capturing central tendencies, variability, and range, providing deeper insights into the dynamics of sensor data. Tools for visualizing sensor signals before and after noise cleaning are included, enhancing the validation of filtering effectiveness and data analysis capabilities. 

This project centers on the Hexoskin Smart Vest, a sophisticated wearable technology designed for Human Activity Recognition (HAR). It focuses on the collection and analysis of biometric data, specifically heart rate, breathing rate, and minute ventilation, which are essential for monitoring various physiological metrics. This smart garment plays a crucial role in advancing healthcare research by providing continuous and real-time data essential for patient monitoring and disease management.

The key contributions of this project include establishing a comprehensive data collection framework that ensures compatibility of the biometric data with machine learning models. By enabling real-time data collection, the Hexoskin Vest significantly enhances patient care in numerous scenarios including long-term health monitoring, clinical trials, and rehabilitation processes. The data collected are instrumental in continuous health monitoring, enhancing patient management, and supporting early detection and preventative healthcare initiatives.

Future work for this project will focus on refining the data collection and analysis processes to enhance integration and utilization within machine learning pipelines. Proposed enhancements include advanced data cleaning methods, innovative feature engineering, and sophisticated data transformation techniques to improve the predictability of health conditions. Additionally, efforts will be made to expand the integration of Hexoskin data with electronic health records to enhance holistic patient care management.