Data Preprocessing Techniques for AI and Machine Learning Readiness: Scoping Review of Wearable Sensor Data in Cancer...

Area of Science:

• Digital Health
• Biomedical Engineering
• Data Science in Oncology

Background:

• Wearable sensors offer continuous patient monitoring in healthcare, particularly in cancer care.
• Challenges persist in ensuring the quality and consistency of data from wearable sensors.
• Preprocessing pipelines for raw sensor data require further optimization.

Purpose of the Study:

• To conduct a scoping review of preprocessing techniques for raw wearable sensor data in cancer care.
• To identify methods preparing data for artificial intelligence and machine learning (AI/ML) applications.
• To understand current approaches for data cleaning, transformation, and feature extraction.

Main Methods:

• Systematic literature search across IEEE Xplore, PubMed, Embase, and Scopus.
• Inclusion criteria: mobile health/wearable sensor studies in cancer, English, Jan 2018-Dec 2023, full text, peer-reviewed.
• Analysis of 20 selected studies meeting eligibility criteria.

Main Results:

• Data transformation (60%) was the most common preprocessing category, converting raw data into informative formats.
• Data normalization/standardization (40%) and data cleaning (40%) were also frequently used to improve data comparability and reliability.
• Techniques addressed noise, missing values, outliers, and feature extraction for AI/ML readiness.

Conclusions:

• High-quality wearable sensor data is crucial for AI/ML in cancer care, but standardized preprocessing lacks.
• A pressing need exists for uniform data quality and preprocessing workflows for wearable sensor data.
• A general framework for preprocessing wearable sensor data is proposed, adaptable beyond cancer care.