Hyperparameter Optimization of Features Selection at the Multiclass Text Classification Stage Using TCO Model in Job Competency Identification

This research proposes an optimized framework named the TCO (Text Classification Optimization) model, which synergizes hyperparameter tuning and feature selection to enhance multiclass text classification tasks, specifically for automated competency identification in recruitment systems. The framework employs Term Frequency–Inverse Document Frequency (TF-IDF) for robust feature extraction, Chi-Square statistical testing for relevant feature selection, and Optuna for efficient hyperparameter optimization. This integrated approach addresses the challenges of high-dimensionality and model generalization in textual datasets. To further improve classification performance, the TCO model implements an ensemble stacking strategy that combines the strengths of multiple base classifiers, including Support Vector Classifier (SVC), Naïve Bayes, Random Forest, Gradient Boosting, and XGBoost. Experimental results show that the combination of optimized hyperparameters and selected features significantly boosts model performance in comparison to baseline models without such enhancements. The TCO model achieved an overall classification accuracy of 74%, with notable precision (86%) in identifying the "Problem Solving" competency class. Balanced recall and F1-score metrics across classes indicate consistent performance and reduced bias. This research contributes meaningfully to the development of fair, interpretable, and scalable automated recruitment systems. By leveraging data-driven methods and ensemble learning, the TCO model offers a reliable solution for organizations seeking to streamline talent acquisition processes while ensuring classification accuracy and competency alignment.