Role of Feature Diversity in the Performance of Hybrid Models—An Investigation of Brain Tumor Classification from Brain MRI Scans

<b>Introduction</b>: Brain tumor, marked by abnormal and rapid cell growth, poses severe health risks and requires accurate diagnosis for effective treatment. Classifying brain tumors using deep learning techniques applied to Magnetic Resonance Imaging (MRI) images has attracted the attention of many researchers, and specifically, reducing the bias of models and enhancing robustness is still a very pertinent active topic of attention. <b>Methods</b>: For capturing diverse information from different feature sets, we propose a <i>Features Concatenation-based Brain Tumor Classification (FCBTC) Framework using Hybrid Deep Learning Models.</i> For this, we have chosen three pretrained models-ResNet50; VGG16; and DensetNet121-as the baseline models. Our proposed hybrid models are built by the fusion of feature vectors. <b>Results</b>: The testing phase results show that, for the FCBTC Model-3, values for Precision, Recall, F1-score, and Accuracy are 98.33%, 98.26%, 98.27%, and 98.40%, respectively. This reinforces our idea that feature diversity does improve the classifier's performance. <b>Conclusions:</b> Comparative performance evaluation of our work shows that, the proposed hybrid FCBTC Models have performed better than other proposed baseline models.