Search Results for feature-importance-
Abstract
Breast cancer is one of the most prevalent and life-threatening diseases worldwide, making early and accurate diagnosis essential for improving patient outcomes. This study aims to evaluate the performance of two machine learning models, Extreme Gradient Boosting (XGBoost) and Gaussian Naive Bayes, for breast cancer classification using the Breast Cancer Wisconsin Diagnostic dataset. A comparative analysis was conducted based on several performance metrics, including Accuracy, Precision, Recall, F1-score, Specificity, Receiver Operating Characteristic (ROC) curve, Area Under the Curve (AUC), Balanced Accuracy, and Confusion Matrix. In addition, feature importance analysis was performed to identify the most influential variables contributing to the classification process. The experimental results demonstrated that both models achieved high classification performance. However, the Gaussian Naive Bayes model showed slightly superior performance, achieving an Accuracy of 96.5% and an AUC of 0.997, compared with 96.0% and 0.993, respectively, for the XGBoost model. Furthermore, the analysis revealed that features related to tumor morphology, particularly concave points mean, perimeter worst, and radius worst, were among the most influential predictors of breast cancer classification. The findings indicate that both machine learning models provide reliable and effective tools for supporting early breast cancer diagnosis, with Gaussian Naive Bayes demonstrating a slight advantage for the dataset used in this study. These results highlight the potential of machine learning techniques to enhance diagnostic accuracy and support clinical decision-making.
Abstract
This study presents a machine learning-based framework for predicting and analyzing student academic performance using behavioral, psychological, demographic, and digital learning features. The dataset consists of 14,003 student records collected from higher education institutions in 2024 and obtained from Kaggle. It includes 16 variables covering study behaviors, engagement patterns, learning resources, psychological factors, and demographic characteristics. The target variable (response) was Final Grade, which was categorized into four performance levels: Low / Fail / Poor performance, below average performance, Good / Average performance, and Excellent / High performance, representing overall academic performance. Three machine learning algorithms XGBoost (Extreme Gradient Boosting), CatBoost (Categorical Boosting), and Support Vector Machine (SVM) were applied to classify student performance levels. The models were evaluated using confusion matrices, accuracy, F1-score, MCC, R², RMSE, MAE, and feature importance analysis. Results show that XGBoost achieved the highest performance (99.96% accuracy), followed by CatBoost (99.82%) and SVM (98.96%). Analysis of the importance analysis revealed that Exam Score is the most influential predictor across all models. The study highlights the effectiveness of advanced machine learning techniques in predicting academic success and supporting data-driven educational decision-making.