Building a Predictive Model for Gynecologic Cancer Using Levels of Data Analytics

Abstract

The four levels of data analytics techniques (descriptive, diagnostic, predictive, and perspective) were used as a methodology. We also used data mining techniques to predict Gynecologic cancer before any lab test or surgical intervention. Influencing and associating between factors are used to cover hidden relationships or unknown patterns. We focused on three types of Gynecologic cancer (cervical, endometrial, and ovarian cancer). We collected an initial examination of 513 (228 benign and 285 malignant) patients from King Abdulaziz University Hospital (Saudi Arabia). Data were collected during the period of 16 years (2000-2016). After examining many models, we found that the classification trees C5 and CHAID beside the Support Vector Machine (SVM) algorithm give the highest accuracy, with the values of 87.33 %, 79.53%, and 78.36 % respectively. The sensitivity and specificity were found to be 86.18 % and 89.00 % for C5. The corresponding values for CHAID were found to be to equals to 82.20 % and 76.72 % while for support vector machine (SVM) the values were found to be 83.74 % and 77.10 %.