Fraud detection in healthcare claims using machine learning: A systematic review

Area of Science:

• Health Informatics
• Machine Learning Applications
• Fraud Detection

Background:

• Healthcare fraud represents a significant financial loss, estimated at 3%-10% of total expenditures.
• Machine learning (ML) offers powerful tools for identifying fraudulent activities in health insurance claims.
• A systematic review is needed to analyze ML techniques applied to this domain over the past two decades.

Purpose of the Study:

• To conduct a systematic literature review of machine learning techniques for health insurance fraud detection.
• To analyze methodologies, data sources, and research trends over the last 20 years.
• To identify current challenges and future opportunities in the field.

Main Methods:

• Searched major academic databases (e.g., Google Scholar, PubMed, IEEE Xplore) for relevant studies.
• Included articles presenting experimental results of ML-based approaches on healthcare claims data.
• Selected 137 articles for qualitative and quantitative analysis.

Main Results:

• A surge in ML-based fraud detection research, with a focus on provider fraud.
• Supervised learning methods (94 studies) are most common, followed by unsupervised (41) and hybrid (12) approaches.
• Traditional ML dominates, but deep learning adoption is increasing; most studies use public datasets, primarily from the US.

Conclusions:

• Key challenges include data inconsistency, lack of standardization, privacy concerns, and limited labeled data.
• Future research should prioritize data transparency, sharing of investigation outcomes, and benchmark datasets.
• Advancements in data sampling, feature encoding, and deep learning are crucial for improving fraud detection efficacy.