Cervical Cancer in Early Stage and Significant Risk Factor Analysis Employing Machine Learning Based Approach

Abstract

Cervical cancer (CC) is the most prevalent and second major reason of death of mortality in women in third world nations when compared to certain other vaginal cancers. It is curable if caught in its early stages. From that standpoint, the study aims to develop an appropriate predictor and computer models for detecting CC at a preliminary phase. Cervical cancer detection in the clinic is extremely expensive. No one wants to go for a clinical test when they have cervical cancer in its early stages. As a result, Machine Learning detection is extremely beneficial. The technology we suggest will detect cervical cancer at an early stage and at a reasonable cost. A CC dataset is compiled with four class attributes such as biopsy, cytology, hinselmann, and schiller, and the dataset is divided into four groups based on target attributes. The dataset was prepared in the data preprocessing phase for better analytical result in the further analysis. Then we applied statistical and EDA approach to discover hidden knowledge from the dataset. To develop machine learning model, different supervised machine learning algorithm like Decision Tree Classifier, Logistic Regression, XG Boost, Multilayer Perception and Random Forest are applied to the dataset to find an efficient classifier. Then, the performances of all the applied classifiers are compared based on accuracy, precision, recall, sensitivity, fmeasure, AUROC, and kappa statistics. We found that RF provided the best performance for biopsy with 94.57% accuracy. MLP and LR generated 98.06% accuracy as the best performing classifier for cytology. Besides, MLP and XGB generated the best performance for hinselmann with 96.51% accuracy, where MLP produced the best result with 94.53% accuracy. Then, we applied four FST methods to rank and show the feature importance for the target feature. Overall, the findings of the study specifies that the proposed model is highly potential to detect CC in early stage.