Predictive modeling for diagnosis of cervical cancer with feature selection

Cervical cancer is one of the most common causes of women's death, especially in low-income communities. The challenge in cervical cancer diagnosis is the absence of symptoms in its early stages. Hence, it is important to identify the risk factors of the disease. This study introduced a data-driven framework for classifying cervical cancer patients based on a set of identified risk factors of demographic information, habits and medical history. This paper applied the supervised machine learning technique of Support Vector Machine (SVM) with three different data pre-treatment processes using Stratified K-Folds cross-validation. In addition, the algorithms of Learning Vector Quantization (LVQ), Recursive Feature Elimination (RFE), and Boruta wrapper were implemented to select the most relevant risk factors. These techniques were applied to a benchmark dataset for cervical cancer risk factors. The results were compared with other studies from the literature which used the same data. The outcome of this study shows more accurate results with 98% prediction accuracy.