Case Study on Effectiveness of Supervised Learning Classifiers for Brain Tumor Detection Using Feature Optimization Techniques

The core objective of this paper is to demonstrate the effectiveness of supervised learning classifiers for brain tumor detection using feature reduction techniques for improving the performance of an underlying pattern analysis task using an appropriate classifier. In the present scenario, tumor growth is becoming a critical concern, and if it is not detected in its early stages, it can be fatal. In this paper, the most prominent feature reduction techniques were used on standard benchmark dataset from Kaggle on brain tumor comprising 3762 rows and 15 feature columns we used PCA (Principal component analysis), <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$t$</tex>-sne for feature reduction to reduce the computational complexity, with increase in accuracy of model. These two techniques which are used in this PCA is used for linear dimensionality reduction whereas for more complex nonlinear patterns, further feature reduced data is fed to distinct supervised learning classifiers like support vector machine, random forest and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$k$</tex> nearest neighbor (KNN)where their accuracy are compared and based on experiments the KNN model outperformed for detection of brain tumor because of its non-parametric memory based learner which avoids overfitting risks by design and makes predictions on basis of local proximity in feature space not on learned fixed model andit is shown using, F1score, recall, precision and plots for visualization.