Machine learning based ensemble approach for cancer detection and classification with explainable AI techniques

This paper presents a state-of-the-art ensemble-based machine learning model of cancer classification based on high dimensionality of gene expression data. Correlation Feature Selection (CFS) is first used to remove redundancy and Salp Swarm Optimization (SSA) is then used to further narrow down the feature space. It has been tested on six publicly available cancer microarray data, i.e., lung, colon, prostate, breast, leukaemia, and brain cancer. Multiple classifiers such as Support Vector Machine, Multi-layer Perceptron, Logistic Regression, Decision Tree, Extreme Learning machine and AdaBoost are examined together with the strategies of bagging and boosting the ensemble. These findings reveal that SSA has greatly enhanced predictive performance and has dramatically decreased the count of the genes chosen. Ensembles based on boosting are always more accurate and robust than bagging on all data sets. Moreover, Explainable Artificial Intelligence (XAI) methods uncover the significant contributions of individual genes to the model decisions giving clear and clinically significant information. The suggested framework has a high potential of reliable and interpretable cancer diagnosis.