eng Oriental Scientific Publishing Company Biomedical and Pharmacology Journal 0974-6242 2026-03-20 19 1 435 446 10.13005/bpj/3363 69958 article Shruthi Mittemari Lakshminarayan Jagannath 1 Nishita Nitin Joshi 2 Tanushree Giridharan 1 Nirmala Devi Manickam 1 Department of Electronics and Communication Engineering, PES University, Bengaluru, India Centre for Healthcare Engineering and Learning, PES University, Bengaluru, India In real-world scenarios, clinical trials provide imbalanced datasets and limited labelled data for the diagnostic evaluation of various medical conditions. Especially in detection of Diabetes Mellitus (DM), which is globally prevalent, the challenge becomes multi-fold in resource-constrained environments. To address these challenges, the study proposes a hybrid framework combining Condensed Nearest Neighbor (CoNN) with Few-Shot Learning (FSL), designed to improve detection speed and reduce memory usage without compromising diagnostic performance. Using publicly available datasets, the framework’s performance was compared with multiple Machine Learning (ML) approaches with an emphasis on preprocessing techniques such as imputation, oversampling, and feature reduction. Compared to conventional models, the proposed CoNN-FSL framework used 1/15^th of the total samples with 2.5 times improvement in terms of training speed. The study offers a comprehensive evaluation of strategies, enhancing the training speed and reducing the storage requirements. Together, these advancements make Machine Learning models more practical and scalable for real-world clinical applications. https://biomedpharmajournal.org/vol19no1/type-2-diabetes-mellitus-detection-using-heterogeneous-machine-learning-models-a-cross-sectional-study/ Condensed Nearest Neighbor Cross-Sectional Studies Data Preprocessing Diabetes Mellitus Few-Shot Learning Machine Learning