Brain tumors form when groups of abnormal cells develop in the brain and have the capacity to infiltrate nearby tissues. Early detection of brain tumors is essential for treating cancer patients and maximizing their survival rates. The brain tumor segmentation (BraTS – 2020) dataset is utilized in this research for segmentation and classification. Min-max normalization and median filter are used in this experiment for data pre-processing after which, the pre-processed data is then fed to DenseNet-201 for extracting features from magnetic resonance images (MRI). Next, a whale optimization algorithm (WOA) is used for effective selection of features. This work proposes an attentive symmetric auto-encoder (ASA)-based segmentation that returns similar code for two variants, and a long short-term memory (LSTM) method for effective classification. The performance of the proposed ASA-LSTM method is estimated by utilizing various tumor regions known as tumor core (TC), enhancing tumor (ET) and whole tumor (WT). The proposed method achieves accuracies of 99.48%, 99.44%, and 99.32% for TC, ET, and WT tumor regions, respectively. These results compared with other existing methods, including convolutional neural network (CNN), artificial neural network (ANN), and recurrent neural network (RNN). The proposed method is found to be effectively than other existing techniques in the segmentation and classification of brain MRI images.