PREDECTIVE ANALYTICS FRAMEWORK FOR INTELLIGENT DECISION SUPPORT SYSTEM TO DIABETES PREDICTION USING ACTIVE LEARNING

Abstract: — currently, medical diagnoses of diseases based on Artificial Intelligence (AI) and Machine
Learning (ML) algorithms are more interesting and required for rapid and precise detection. Furthermore, the
Intelligent Decision Support System (IDSS) of healthcare systems using big data analytics is more helpful and
effective. The utilization of Active Learning (AL) approaches is more crucial to enhance the diagnosis process
for diabetes by integrating the experience and feedback of human–experts with lightly available labeled data. In
this paper, we presented multi-labeled active learning algorithms with both classification and selection strategies
to detect diabetes disease. Moreover, the proposed framework utilized four active learning algorithms called
AUDI, RANDOM, MMC, and Adaptive. These AL algorithms are promising for multi-label selection of
diabetes because of their ability to reduce the cost of querying the labeled selected data based on AL strategies.
The evaluation results of the proposed AL framework prove the ability to classify diabetes with and without
optimization of hyper-parameter values based on a grid search algorithm. The result indicated that for the
optimal label ranking model, the selection approach is used over others due to accuracy in generalization of the
learning model beyond the current data. In terms of the Recall without grid search optimization parameters,
however, the selection technique was highlighted.