Computational intelligence in early diabetes diagnosis: a review

Area of Science:

• Computational intelligence
• Machine learning
• Medical informatics
• Diabetes diagnosis

Background:

• Effective diabetes diagnosis systems using computational intelligence are a current research priority.
• Numerous machine learning and artificial network approaches have shown high accuracy (up to 99%) on diabetes datasets, primarily from Pima Indian individuals.
• Despite high predictive accuracy, these computational tools have not yet achieved widespread clinical application.

Purpose of the Study:

• To outline the diverse options, recent advancements, and potential of machine learning algorithms for diabetes diagnosis.
• To inform diabetologists and clinical investigators about computational diagnosis tools.
• To highlight supervised and unsupervised methods impacting diabetes detection and diagnosis.

Main Methods:

• Review and explanation of various machine learning algorithms for diabetes diagnosis.
• Focus on supervised and unsupervised learning techniques.
• Discussion of theoretical advancements in algorithmic construction, learning theory, generalization performance, and incorporation of prior knowledge and uncertainty.

Main Results:

• Identification of accurate machine learning algorithms for diabetes diagnosis.
• Analysis of the advantages and disadvantages of different methodologies.
• Emphasis on algorithms showing promise for enhancing diabetes diagnosis accuracy and applicability.

Conclusions:

• Machine learning holds significant potential for improving diabetes diagnosis, but clinical integration remains a challenge.
• A deeper understanding of algorithmic theory and practical considerations is crucial for successful implementation.
• This review serves as a resource for researchers to expand knowledge in computational intelligence-based diabetes diagnosis.