Exploring Machine Learning Approaches for Predicting Brain Tumors

Abstract

Brain tumours present a significant health challenge worldwide, necessitating accurate and timely prediction for improved patient outcomes. ML techniques have emerged as promising tools for brain tumour prediction, leveraging their ability to learn from vast datasets and identify complex patterns within medical images. In this paper, we conduct a comparative analysis of various machine learning algorithms to assess their effectiveness in predicting brain tumour presence and classifying tumour types based on medical imaging data. The dataset used in this research consists of a diverse collection of brain MRI scans, encompassing both tumour-afflicted and healthy brain samples. We preprocess the data to ensure uniformity and feature extraction, which includes texture analysis, intensity histogram, and shape descriptors, to represent the regions of interest effectively. In this research endeavor, we assess and juxtapose eight distinct machine learning algorithms. The algorithms subjected to scrutiny encompass SVM, Random Forest, Decision Tree, K-Nearest Neighbors, and Cat Boost. Each of these algorithms undergoes comprehensive training and meticulous testing using a preprocessed dataset. The primary objective of this evaluation is to gauge their predictive capabilities across a spectrum of performance metrics, which includes accuracy, sensitivity, specificity, and the area beneath the ROC curve. The results indicate that all algorithms demonstrate respectable prediction capabilities, with SVM and Neural Networks exhibiting the highest accuracy and sensitivity. However, the performance of each algorithm varies concerning computational efficiency and interpretability. SVM proves to be efficient and reliable in predicting tumour presence, while Neural Networks offer robustness in predicting tumour subtypes. In conclusion, this study provides an exhaustive comparative analysis of ML algorithms for brain tumour prediction. The findings contribute valuable insights into the strengths and weaknesses of each algorithm, guiding medical practitioners and researchers in selecting appropriate techniques for brain tumour classification and advancing the field of medical image analysis for improved brain tumour diagnostics. Future research should focus on addressing the identified challenges and exploring ensemble methods to further enhance the accuracy and reliability of brain tumour prediction models.