<p>Brain tumor (BT) is the deadliest disease developed due to the proliferation of abnormal glial cells and it is curable only if the tumor is detected at the initial stage. It is also significant to know whether they are benign (noncancerous) or malignant (cancerous). Through magnetic resonance imaging (MRI) images and computerized tomography (CT), the prediction is possible, manual prediction may lead to inaccurate prediction and is also time consuming. Therefore, to avoid such drawn-out and time-consuming methods, Deep Learning (DL) models are used instead but still, they hold some limitations. Therefore, this research presents an effective tumor segmentation and prediction utilizing the probability theory and White-Headed Timber Optimization-based Deep CNN model (WHT-based Deep CNN) model that provides precise BT prediction from MRI scans. The proposed segmentation approach is exhibited effectively by employing the Deep flow-based optimized tumor segmentation that utilizes the probability theory for obtaining the precise segmentation of the tumor region (TR). Further, the optimal centroid values determined with the WHT optimization assist in locating the exact TR. Hence, the refined features extracted from the segmented region are classified using the optimized deep CNN classifier, which is constructed with eight convolutional layers. The model training is supported with adaptive parallel learning that selects the optimal Deep CNN parameters for acquiring the maximal detection accuracy. The WHT-based Deep CNN is outperformed as 97.46% accuracy, 97.22% sensitivity, and 97.89% specificity for the Brats 2019 dataset that surpassed the performance of other existing techniques in BT prediction.</p>