Preemptive Diagnosis of Colorectal Cancer Using Computational Intelligence Techniques

One of the most common cancers in the world is colorectal cancer. Due to this increasing global concern, the focus of this study was to enhance the predictive power of Machine Learning (ML) techniques for making preemptive diagnoses of colorectal cancer. This paper uses an online dataset to develop models for colorectal cancer and employs three ensemble ML algorithms, Histogram-based Gradient Boosting (HGB), Adaptive Boosting (AdaBoost), and Extra Trees (ET). The dataset comprised 115 instances, each with 11 features, after it underwent rigorous pre- processing including, imputing null values and one-hot encoding, followed by the utilization of the Synthetic Minority Over-sampling Technique accompanied by Edited Nearest Neighbors (SMOTE- ENN). Another key aspect of this study is the optimization of these models, achieved using GridSearchCV coupled with 5-fold-cross validation to identify the values of the most effective hyperparameters. A notable breakthrough is the enhanced performance of the HGB and Adaboost models in comparison with previous studies, as both models achieved a perfect 100% accuracy, precision, recall, and F1 score higher than previous studies where the F1 was about 88%. This outcome demonstrates the efficiency of using computational intelligence, specifically ML, in revolutionizing the preemptive diagnosis of colorectal cancer, offering a promising platform for future research and clinical application.