Classification of Stroke Desease Using the Learning Vector Quantization Algorithm

Abstract

Stroke is one of the leading causes of death and disability worldwide, thereby making early detection crucial for timely and appropriate medical treatment. In clinical practice, stroke diagnosis is generally carried out through medical examinations and patient history analysis, but this process is time-consuming and depends on the subjective judgment of medical personnel. Therefore, machine learning approaches can be utilized to support disease classification more quickly and objectively. This study aims to analyze the performance of the Learning Vector Quantization (LVQ) method in classifying stroke disease using a dataset obtained from Kaggle. The dataset used in this study is imbalanced;therefore, the SMOTE (Synthetic Minority Over-sampling Technique) method was applied to handle class imbalance. The research stages included data preprocessing, splitting data into training and testing sets, LVQ model training, parameter optimization using learning rate and maximum epoch, and model evaluation using accuracy and sensitivity. The results show that the LVQ model trained on the original dataset achieved an accuracy of 95,72%, but failed to detect stroke cases with a sensitivity of 0%. After applying SMOTE, the best model achived a stroke sensitivity of 90%, although the accuracy decreased to 49,49% due to the high number of false positives. These findings indicate that LVQ is highly sensitive to data distribution and model parameters, making its performance on this dataset less optimal for stroke classification and more suitable as an initial screening tool.