Machine Learning Models and Applications for Early Detection.

Area of Science:

• Computer Science
• Data Science
• Artificial Intelligence

Background:

• Early detection (ED) of anomalies is critical for timely decision-making and mitigating negative impacts.
• Machine learning (ML) offers powerful tools for developing anomaly detection systems.
• This review focuses on ML models for ED, particularly within fraud detection applications.

Purpose of the Study:

• To conduct a literature review of ML models used for early anomaly detection.
• To analyze how these models function in a multidisciplinary context, with a specific focus on fraud detection.
• To categorize ML models into Single Base Models (SBMs) and Stacking Ensemble Models (SEMs).

Main Methods:

• Literature review of multidisciplinary research on ML for ED.
• Categorization of ML models into SBMs and SEMs.
• Analysis of reported accuracy metrics for various ML models.

Main Results:

• Multiple ML models, including Logistic Regression, SVMs, Random Forests, and XGBoost, are effective for ED.
• SBMs achieved accuracies over 80%, while SEMs surpassed 90% in general ED tasks.
• MLMs in fraud detection consistently reported accuracies exceeding 90%.

Conclusions:

• ML models provide a robust and accurate method for identifying and classifying anomalies.
• MLMs are highly effective for early anomaly detection in fraud, processing large datasets efficiently.
• The application of MLMs in fraud detection aids in preventing financial losses through rapid detection of suspicious activities.