Research Square (Research Square), Год журнала: 2025, Номер unknown

Опубликована: Март 28, 2025

In the field of machine learning for diabetes detection, outliers in datasets remain a significant challenge. Traditional outlier handling methods often fall short terms accuracy and are prone to Type II errors. Moreover, these conventional approaches typically discard outliers, leading inefficient data utilization. To address limitations, this study aims develop more effective unsupervised detection mechanism by integrating Conditional Generative Adversarial Networks (CTGAN) with Autoencoders. We further introduce secondary layer based on Outlier Factor enhance reduce Additionally, we incorporate into an ensemble framework propose novel training method base learners that retains rather than discards outliers. The resulting model architecture is capable simultaneously performing classification tasks. Our demonstrates exceptional performance eight three datasets. Ablation studies confirm proposed dual effectively mitigates Experimental results show that, compared traditional methods, approach achieves improvements accuracy, reduction errors, enhanced utilization efficiency models.

Язык: Английский