Implementation of K-Prototypes with Feature Selection in Clustering Cervical Cancer Patients based on Risk Factors

Cancer is a leading cause of death worldwide, resulting in nearly 10 million deaths or almost one-sixth of all deaths in 2020. Effective primary prevention measures can prevent at least 40% of cancer cases. Cancer mortality rates are higher in developing countries than in developed countries, reflecting disparities in addressing risk factors, detection success, and available treatments. Women in developing countries most frequently suffer from cervical cancer. It is crucial for communities, especially women, to have knowledge about the risk factors for cervical cancer. One potential solution to this issue is the role of machine learning in analyzing cervical cancer patient data. This study uses the K-Prototypes clustering algorithm, which can cluster mixed data, both numerical and categorical. Cervical cancer risk factor data were used in this research. Feature selection was performed to improve the performance of the K-Prototypes algorithm, using feature selection methods Variance Threshold and Correlation Coefficient. The best performance of the K-Prototypes algorithm was obtained using the Correlation Coefficient, as reviewed based on a Silhouette Coefficient of 0.6, a Davies-Bouldin Index of 0.6, and a Calinski-Harabasz Index of 1.080. Interpretation of the clusters formed revealed major differences in the characteristics of risk factors between two clusters, namely age, menopause, and health conditions such as leukorrhea, bleeding, lower abdominal pain, and loss of appetite. Meanwhile, factors related to previous history, reproductive health, and nutritional issues did not show significant differences. The K-Prototypes algorithm is expected to be a solution in identifying groups based on cervical cancer risk factors to assist medical professionals in decision-making and subsequent actions, as well as to provide knowledge to the public.