Enhanced Classification of Ear Disease Images Using Metaheuristic Feature Selection

Ear diseases are characterized by various symptoms, including balance disturbances, delayed speech development in children, headaches, fever, and hearing loss. To prevent further complications, these conditions must be diagnosed and treated promptly. The traditional diagnostic method has been an otoscope examination by otolaryngologists. However, the accuracy of this approach is contingent upon the clinician's expertise and the quality of the equipment used, which can render it susceptible to misdiagnosis. Incorrect diagnoses may result in the administration of antibiotics unnecessarily, disease progression, and other adverse consequences. This study aims to evaluate the efficacy of computationally efficient machine learning models in classifying ear disease images. To enhance classification accuracy, a Histogram of Oriented Gradients (HOG) was employed for feature extraction and optimization algorithms were utilized for feature selection. The Whale Optimization Algorithm (WOA) effectively selected informative features for the k-Nearest Neighbors (kNN) model, achieving a classification accuracy of 92.6%. Furthermore, the Support Vector Machine (SVM) model achieved an accuracy of 92% using a feature map comprising features selected by a range of optimization algorithms. The experimental findings emphasize the potential of strategic feature selection in enhancing the performance of classical machine learning models for ear disease classification. By employing computationally efficient techniques such as HOG and optimization algorithms, these models can attain classification accuracies that are on par with those of more resource-intensive deep learning approaches. Such developments facilitate the creation of accessible and efficient diagnostic tools, particularly beneficial in resource-constrained clinical settings. The findings of this study provide a basis for further research to enhance the diagnostic precision of machine learning-based techniques in medical imaging.