Review of Machine Learning Algorithms for Diagnosing Mental Illness

Area of Science:

• Healthcare technology
• Data science in medicine
• Computational psychiatry

Background:

• Advancements in computing and internet technologies have increased data scale and quality in healthcare.
• Machine learning (ML) is pivotal for analyzing large datasets but often misunderstood in application.
• Misconceptions exist regarding ML capabilities, such as solving small sample size problems or equating deep learning with all ML.

Purpose of the Study:

• To review research on diagnosing mental illness using ML algorithms.
• To suggest practical applications and methodologies for ML techniques in mental health.
• To address common misunderstandings about ML algorithm application in this field.

Main Methods:

• Systematic review of research on mental illness diagnosis using ML.
• Focused on five frequently used traditional ML algorithms: Support Vector Machines (SVM), Gradient Boosting Machine (GBM), Random Forest, Naïve Bayes, and K-Nearest Neighborhood (KNN).
• Organized and summarized the application of these algorithms in mental health research.

Main Results:

• SVM, GBM, Random Forest, Naïve Bayes, and KNN are commonly used in mental health ML research.
• Many studies fail to justify the choice of ML algorithm, despite each having unique advantages.
• Some research applies ML algorithms without fully considering the underlying data characteristics.

Conclusions:

• Researchers must understand the specific properties and limitations of chosen ML algorithms.
• Awareness of data characteristics is essential for appropriate ML application.
• This paper offers insights into ML algorithm properties and limitations for practical use in mental health.