SALTS: Streamlined Adaptive Learning for Sensors Time Series.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Data Science

Background:

• Sensor-generated time series data in healthcare offers significant potential but faces labeling challenges due to sequential nature and temporal dependencies.
• Manual data labeling is costly, and domain experts may lack specialized skills for optimizing machine learning models.
• Automating machine learning model training is crucial for efficient analysis of medical time series data.

Purpose of the Study:

• To develop an automated approach for training machine learning models on medical time series data.
• To enhance the efficiency of healthcare data analysis by optimizing data acquisition and model refinement.
• To reduce the reliance on human input for machine learning model tuning in medical applications.

Main Methods:

• Adaptive data acquisition to select informative samples for labeling.
• Dynamic model refinement to optimize hyperparameters on-the-fly.
• Automatic learning phase to maximize the utilization of unlabeled samples.
• The SALTS (Self-Adaptive Learning for Time Series) strategy integrates adaptive data acquisition and dynamic model refinement.

Main Results:

• The proposed method outperforms existing baselines and state-of-the-art approaches in classifying EEG, ECG, and IMU health signals.
• Demonstrated significant reduction in the need for human input during model tuning.
• Enhanced the efficiency of machine learning application in healthcare time series analysis.

Conclusions:

• The SALTS method provides a robust learning strategy that continuously refines models with expanding data and human expertise.
• It maximizes the information gained from each human annotation step in an automated manner.
• SALTS enhances the applicability and efficiency of machine learning for healthcare time series data analysis.