Diabetes Classification Problem with CatBoost Method and Optuna Gradient Boosting Optimization

Diabetes mellitus is a dangerous global epidemic that can cause kidney failure, heart attack, blindness, and death. The main cause of this disease is too high blood sugar levels due to the irrelevant work of pancreatic beta cells. Rapid computer technology advances and implementations based on artificial intelligence have penetrated various fields, especially the health sector. Data mining and machine learning are technologies that continue to grow, are reliable, and can be used as research support tools in the medical field. This technology is considered capable of helping to automate diabetes prediction. Based on related research, our contribution to this research is to utilize the CatBoost method with Optuna optimization. In the data preprocessing stage, all 768 data sets have gone through removing outliers, normalizing features to reducing dimensions. The dataset is divided into $85 \%$ composition for training data and $15 \%$ for test data. At the feature selection stage with Correlation Pearson’s, we found that Glucose correlated with diabetes outcome or disease because it had a value range of 0.47. The results’ accuracy is considered less than optimal because it only produces an accuracy value below $80 \%$, even after optimization with Optuna. It can be seen that there was only an increase of $7 \%$ after optimizing with Optuna.