Predictive modeling based on small data in clinical medicine: RBF-based additive input-doubling method.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Machine Learning for Healthcare

Background:

• Handling limited medical datasets is a significant challenge for health decision support systems.
• Accurate classification and regression are vital for numerous medical tasks, especially with scarce data.
• Existing regression methods may struggle with the inherent noise and variability in small medical data samples.

Purpose of the Study:

• To improve the accuracy of regression analysis for short medical data sets.
• To enhance the RBF-based input-doubling method for better performance in medical applications.
• To provide a more reliable tool for health decision support systems dealing with limited observations.

Main Methods:

• Modification of the RBF-based input-doubling regression method by introducing averaging elements.
• Integration of ensemble method principles to compensate for prediction errors of varying signs.
• Experimental validation using a real-world short medical dataset from rheumatology (77 observations).

Main Results:

• The enhanced RBF-based additive input-doubling method demonstrated superior prediction accuracy compared to existing methods.
• Optimal parameters were identified experimentally, maximizing accuracy based on Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE).
• The improved method achieved higher accuracy without an increase in the training algorithm's duration.

Conclusions:

• The proposed RBF-based additive input-doubling method effectively handles short medical data, significantly improving regression accuracy.
• The method's adaptability allows for integration with other artificial intelligence tools for broader medical applications.
• This advancement offers a valuable solution for data-scarce scenarios in medical research and clinical decision support.