Grey Wolf Optimizer-Based Feature Selection for Naive Bayes Lung Cancer Risk Classification

Abstract

Lung cancer risk classification is an important task in health informatics because early identification can support further clinical assessment and risk management. Machine learning can process patient risk factors and symptoms, but classification performance can be affected by irrelevant or less informative input features. This study addresses this problem by applying the GreyWolf Optimizer(GWO)asafeature-selectionmethodforNaiveBayesclassification of three lung cancer risk levels: Low, Medium, andHigh. Thedatasetcontains 1,000 patient records with 25 attributes related to lung cancer risk factors and symptoms. The proposed procedure includes data preprocessing, an 80:20 training-testing split, GWO-based feature selection, and Naive Bayes classification. Each candidate solution in GWO represents a feature subset, while Naive Bayes classification accuracy is used to evaluate the candidate solutions. The baseline Naive Bayes model achieved an accuracy of 89.50%. GWO reduced the number of input attributes from 25 to 10, and the resulting Naive Bayes model achieved an accuracy of 96.50%. Precision, recall, and F1-score also increased across the Low, Medium, and High risk classes after feature selection. These results show that GWO-based feature selection reduced the input dimensionality while producing higher classification metrics than the baseline Naive Bayes model under the evaluated experimental setting.