Application of Generalized Relevance Linear Vector Quantization for Diabetes Diagnosis

Numerous statistical machine learning techniques have been proposed for solving a variety of classification problems. Prototype-based models, such as standard learning vector quantization (LVQ) and its extensions, have been widely applied to various applications domains due to their intuitive nature and simplicity. This paper adopts LVQ and its three variants, namely, generalized learning vector quantization (GLVQ), relevance learning vector quantization (RLVQ), and generalized relevance learning vector quantization (GRLVQ) algorithms for the problem of diabetes disease classification. Four different error metrics were used to measure the robustness and accuracy of each classifier. These measures include root mean squared error (RMSE), mean zero–one error (MZE), mean absolute error (MAE), and macro averaged mean absolute error (MMAE). The obtained results indicate that GRLVQ was very effective, which produced a minimum error in terms of all error metrics used.