Integrating Machine Learning Algorithms and Advanced Computing Technology Using an Ensemble Hybrid Classifier

Abstract

Ovarian Cancer (OC) is one of the major types of cancers in women worldwide. Despite the standardization of characteristics that can help distinguish benign from malignant ovarian masses, accurate predictive modelling following ultrasound (US) examination and biomarkers for ’progression-free survival’ is lacking in the field of ovarian cancer. Important leading factors in ovarian cancer lethality are the lack of diagnostic procedures and proper screening to detect early-stage ovarian cancer, and the rapid spread of the disease over the surface of the peritoneum. Therefore, developing tools for accurate screening and prognosis, as well as the diagnosis of early stage ovarian cancer, is a current clinical need. In this study, an ensemble classifier was developed as a novel means of ovarian cancer prediction, and its effectiveness was assessed. The ensemble classifier integrates various machine learning algorithms, including support vector machines (SVM), k-nearest neighbors (KNN), decision trees (DT), naïve Bayes (NB), and logistic regression (LR). Because ensembles may integrate the benefits of numerous models, they can mitigate the limitations of each model individually and improve the overall predictive performance, making them popular in the domain of machine learning. To increase predictive performance, an ensemble hybrid approach was created by utilizing a meta-classifier to merge many base classifiers. The performance with respect to various measures of the ensemble classifier was evaluated considering a comprehensive novel dataset of ovarian cancer patients, including tumor markers as well as clinical and ultrasound features. Through extensive cross-validation studies, the hybrid model showed better prediction accuracy of 95% which is approximately 6-17% improved than the baseline classifiers and state-of-the-art ensemble approaches in predicting ovarian cancer. After comparing the performance of the ensemble classifier with other existing classifiers, the ensemble classifier outperformed the individual models and conventional diagnostic techniques in terms of sensitivity (94%) and specificity (95%) through performance evaluation.