Optimizing text classification: Particle swarm approach to random forests

The approach proposed in this paper is a hybrid text classification design that ensembles the strengths of Random Forests and Particle Swarm Optimization. PSO is used to optimize the crucial hyperparameters of the base model-Random Forest classifier, like the number of estimators and maximum depth, it enhances the accuracy and sturdiness of the classifier. The 20 Newsgroup dataset was pre-processed via TF-IDF vectorization, and performance was tested with the help of accuracy value, precision-recall, macro F1-score and visualized using ROC-AUC curves. The model results proved that the PSO-optimized base model outperformed the baseline Random Forest, in terms of both accuracy and F1-score. An integrated analysis of features subset was done, revealing the understandability of key attributes. This ensembled model proposed here is a scalable and graspable resolution for text classification tasks involving high dimensional data.