Enhancing Brain Tumor Diagnosis: A Hybrid Machine Learning Model with PCA and SHAP for Interpretability

To enhance patient prognosis, identifying and diagnosing brain tumors at an early stage are significantly important. For this purpose, we present a hybrid machine learning based framework for brain tumor identification which utilizes principal component analysis (PCA), support vector classifiers (SHC), and Shapley additive explanations (SHAP) for optimal performance and interpretability. In medical imaging data, the degree of complexity and variety is so high that in order to improve efficiency and minimize overfitting, it is more reasonable to employ dimensions reduction techniques which as support vector classifiers (SVC) PCA to SVC is used in Shapley to increase computation speed without loss of important diagnostic features. Because of its robustness to high-dimensional data and ability to provide clear decision boundaries, SVC is used as the primary classifier, while logistic regression is employed as the benchmark for comparison. To enhance the explainability of machine learning healthcare models, SHAP is used to understand the role different predictors play which will in turn increase clinicians trust in the model. The approach is aimed to the Brain Tumor MRI Dataset at Kaggle, where the model has claimed to have outperformed others. The model outperforms the baseline model in terms of accuracy, precision and recall when identifying cases with and without a brain tumor. Through the integration of dimensionality reduction and interpretable machine learning techniques, we not only achieve accurate diagnostic results, but also address the issue of reconciling sophisticated algorithmic models with clinical use. This approach serves as a potential real-time brain tumor diagnosis and patient management tool.