Dimensionality Reduction and Gradient Boosting for In-Vivo Hyperspectral Brain Image Classification

Hyperspectral imaging (HSI) datasets contain hundreds of contiguous narrow spectral bands, which can create challenges in data analysis due to the curse of dimensionality. However, much of this data is redundant, necessitating the application of dimensionality reduction techniques to extract the most relevant features before the classification task. Two feature extraction methods, principal component analysis (PCA) and non-negative matrix factorization (NMF), are explored. Then, the reduced feature sets are used to train three gradient-boosting algorithms, namely extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and categorical boosting (CatBoost), for the classification of in-vivo hyperspectral brain images. The experimental results show the effectiveness of gradient boosting over support vector machine (SVM) and random forest (RF) in performance measures like accuracy, sensitivity, specificity, and F1-score.