Optimasi Model Particle Swarm Optimization (PSO) Menggunakan SMOTE Untuk Menentukan Penyakit Diabetes Mellitus

Abstract

Diabetes mellitus is a chronic disease that continues to increase globally and can affect various age groups. If not properly managed, this disease can lead to serious complications. In recent years, technological advancements, particularly in the field of machine learning, have significantly contributed to improving the accuracy of diabetes diagnosis and prediction. This study utilizes the Decision Tree algorithm, enhanced by two optimization methods: the Synthetic Minority Over-sampling Technique (SMOTE) to address data imbalance and Particle Swarm Optimization (PSO) to optimize the model's hyperparameters, thereby improving classification accuracy. The dataset used in this study is the Diabetes Prediction Dataset available on Kaggle, consisting of 100,000 entries. Based on the analysis results, the implementation of data preprocessing and hyperparameter optimization has proven to increase the model's accuracy from 95.21% to 96.52%. Additionally, an evaluation using the confusion matrix shows an improvement in precision from 70.82% to 86.19% and an increase in the F1-score from 72.49% to 78.52%, although there is a slight decrease in recall from 74.24% to 72.11%. These findings demonstrate that a combination of data preprocessing, data balancing, and hyperparameter optimization can significantly enhance the performance of a classification model in detecting diabetes. For future development, it is recommended that the model be tested on other datasets to improve generalizability. Furthermore, exploring additional algorithms such as Random Forest or XGBoost could be beneficial in obtaining more optimal results.