Brain tumor detection becomes a complex task, and detecting the brain tissues is even more difficult for radiologists through the manual process. Due to the complexity of brain anatomy, variability in tumor characteristics, and technical limitations, the detection and classification of tumors leads to a challenging task from the magnetic resonance images (MRI). The deaths are increasing nowadays due to a lack of early awareness of the tumor. The tumor symptoms, like headache, fever, etc., are very normal at the early stage. Generally, in real practice, we ignore and consume medicines without consulting a doctor, which creates a huge problem of unknown diseases. Even a minor mistake could pose a threat to life. Doctors and radiologists detect and classify tumors manually, which takes lot of time for the patients. This paper proposes an automatic detection and classification technique based on Deviation Sparse Fuzzy C-Means (DSFCM) clustering for segmentation and Crow Search Algorithm (CSA) Extreme Learning Machine (ELM) for the classification of brain tumors from magnetic resonance images (MRI). The dataset was considered to be from the Kaggle website for this research purpose. The proposed CSA-ELM model achieved an accuracy of 99.23%, which is better than the conventional models. The detailed mathematical analysis and the results are presented.