Over time, Magnetic Resonance Imaging (MRI) has played a pivotal role in accurately delineating brain cancers, facilitating diagnostic processes for medical practitioners. However, the inherent subjectivity in human interpretation presents challenges, prompting the exploration of machine learning models to augment diagnostic guidance. Despite their advantages, these models can be susceptible to errors depending on their objectives. This study addresses segmentation errors induced by data capture noise by introducing the morphological fractional order Darwinian particle swarm optimization (M-FODPSO) approach. Leveraging classical discrete wavelet transform and principal component analysis, pertinent features are extracted from M-FODPSO outputs, subsequently trained on an ensemble classifier. Performance evaluation demonstrates an impressive accuracy of 97.03%, achieved with an average processing time of 1.7161 s, thereby showcasing enhanced tumor cell characterization compared to the classical FODPSO method. This approach effectively mitigates segmentation errors attributed to data noise, underscoring its potential for refining MRI-based brain cancer diagnosis.