A Heuristic and ANN based Classification Model for Early Screening of Cervical Cancer

Abstract

Cervical cancer is one of the most leading causes of mortality among women worldwide. This deadly disease could be prevented by vaccines and easily cured if detected at an early stage. Various researchers focus on providing methods for unambiguous results of screening tests for early diagnosis of cervical cancer and also on detecting stages of cervical cancer through Pap smear images of the cervix. Various socio-economic factors of women in underdeveloped countries limit the regular Pap smear test for screening of cervical cancer. It is pragmatic that the prediction on the likelihood of cervical cancer is not always possible based on the fewer inquiries from the patients and the data remain inadequate. Oversampling of the data is needed to any dataset for preprocessing the data and this is achieved by using Synthetic Minority Oversampling Technique (SMOTE). In the proposed work, chi-square, a filter-based feature selection method is used to select the attributes based on their correlation between feature and the class to remove the irrelevant attributes from the dataset. Further genetic-based feature selection is used to filter the best optimal features from the selected attributes. Linear Support Vector Machine (SVM) classifier is applied to the selected attributes from the genetic algorithm to aid in predicting the model through training and testing, resulting in an accuracy of 93.82%. Backpropagation, a deep learning method is used as a classification model for cervical cancer, resulting in an improved accuracy of 97.25%. The experimental results show the efficiency of the proposed model is better in comparison to the previous models in terms of accuracy.

Keywords

• Cervical cancer
• SMOTE
• SVM classifier
• Backpropagation
• Deep Learning