<![CDATA[SEO optimization plays a critical role in improving web page visibility in search engines. However, the limited proportion of Good SEO articles creates class imbalance, which biases classification models toward the majority class and reduces their ability to detect high-performing content. This study investigates effective strategies for handling data imbalance in SEO performance classification and evaluates the suitability of tree-based ensemble models for this task. Two oversampling techniques, the Synthetic Minority Oversampling Technique (SMOTE) and Adaptive Synthetic Sampling (ADASYN), were applied to a dataset containing textual and numerical SEO features. Textual components were transformed using FastText embeddings to capture semantic relationships among on-page elements. Random Forest and XGBoost models were then trained and evaluated using balanced accuracy, precision, recall, and F1-score. The experimental results indicate that Random Forest combined with SMOTE achieved the highest balanced accuracy of 74.2%, outperforming other configurations in identifying Good SEO articles. Although XGBoost with ADASYN also achieved a competitive balanced accuracy of 74.0%, its performance remained slightly lower than Random Forest with SMOTE. These findings confirmed that appropriate imbalance handling significantly enhanced classification fairness and minority-class sensitivity. From a computer science perspective, this study advances research in imbalanced text classification and web mining by providing empirical evidence that integrating semantic embeddings with adaptive resampling strategies improves predictive robustness and model reliability in real-world digital content evaluation. The proposed framework supports the development of automated, data-driven SEO auditing systems. ]]>
Copyrights © 2026
