Improving Diagnostic Accuracy on Prescription Text Data Using SMOTE-Optimized SVM

Disease classification based on drug prescription data plays a crucial role in helping healthcare professionals understand patient health conditions and supporting clinical decision-making. Drug prescription data actually contains a wealth of information regarding disease indications, but is generally presented in unstructured, free-text form. Furthermore, the data distribution across disease classes is often imbalanced, with some diseases receiving less data than others. This can lead to inaccurate classification models that favor disease classes with more data. This study aims to enhance the performance of disease classification based on drug prescription data by combining text mining approaches, the Synthetic Minority Oversampling Technique (SMOTE), and the Support Vector Machine (SVM) algorithm. The research process begins with text preprocessing, which includes case folding, tokenization, stopword removal, and stemming, to clean and normalize the prescription data. Next, the text data is converted into numeric features using the Term Frequency–Inverse Document Frequency (TF-IDF) method to enable processing by machine learning algorithms. To address the class imbalance issue, the SMOTE method is applied to training data by generating synthetic data for a limited number of disease classes. A classification model was then built using the SVM algorithm, known to be effective in handling high-dimensional text data. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. The results showed that the application of SMOTE and parameter optimization in Support Vector Machine significantly improved classification performance, with an accuracy of 92.6%, a precision of 91.8%, a recall of 93.4%, and an F1-score of 92.6%. The increased recall value in the class of patients diagnosed with diabetes indicates that the model is able to correctly identify most diabetes cases based on medical prescription data.