An Efficient Weighted Majority Voting Ensemble Machine Learning Classifier Framework for Image Segmentation

Ensemble learning is an effective approach to improving the robustness and accuracy of image segmentation by combining multiple classifiers. This study presents an efficient Weighted Majority Voting Ensemble (WMVE) framework integrating five Machine Learning (ML)-based segmentation models: Random Forest (RF), Naïve Bayes (NB), eXtreme Gradient Boosting (XGB), K-Nearest Neighbors (KNN), and Multilayer Perceptron (MLP). The standard preprocessing techniques include Gaussian smoothing and median filtering to ensure reliable feature extraction and high-quality segmentation. The proposed WMVE allows assigning specific weights to each classifier according to its validation accuracy result for adaptive informed decision fusion, where performance evaluation uses region-based segmentation metrics, including Segmentation Covering (SC), Probabilistic Rand Index (PRI), Variation of Information (VoI), Global Consistency Error (GCE), and Boundary Displacement Error (BDE). Experimental results indicate that the proposed ensemble is better than single classifiers and nearly as good as existing ensemble and deep clustering approaches on several datasets. Therefore, the WMVE framework can be considered a strong approach to attain high performance in image segmentation, since experimental results also show near-optimal performance with existing state-of-the-art methods.