XGBoost Classifier with Hyperband Optimization for Cancer Prediction Based on Geneselection by Using Machine Learning Techniques

摘要

In the medical field, gene selection is critical, and it has the ability to diagnose diseases at an early stage. Data imbalance and poor feature selection performance are limitations in current techniques. Hyperband optimization is proposed in this paper to increase the performance of the XGBoost classifier. The NCBI gene dataset is utilised to evaluate the developed technique's performance in gene selection. The normalization procedure is used to scale the input data and decrease data discrepancies. When the Principal Component Analysis (PCA) method is used on input data to choose important features for classification, the independence variable becomes more difficult to interpret. To execute the gene selection for disease diagnosis, the selected features are applied to the XGBoost classifier. The hyperband optimization method searches in a distributed fashion to increase parameter exploration. The accuracy of the XGB-PCA-HO approach is 97.06%, XGB is 88.24%, and Random Forest is 85.29%.