Predicting endometrial cancer using machine learning

Endometrial cancer (EC) is one of the most common types of cancer among women and has a high mortality rate. In diagnosis, biomarkers such as Cancer Antigen 125 (CA125), a blood parameter, provide significant benefits. Therefore, this study attempted to determine the level of risk for endometrial cancer by evaluating the age, weight, number of children, and CA125 value in the blood of individuals. For this purpose, eleven classical and two alternative machine learning (ML) algorithms were used. Logistic Regression, Naive Bayes, K-Nearest Neighbors, Support Vector Classification, Multi-Layer Perceptron (MLPC), Decision Tree (CART), Random Forest (RF), Gradient Boosting Machines (GBM), XGBoost, LightGBM, and CatBoost are the classical machine learning (ML) algorithms used. Fast Linear Network (FLN) and Extreme Learning Machine (ELM) were also used as alternative algorithms. In this study, CA125 values were obtained from a total of 220 patients, 185 of whom had endometrial cancer and 35 of whom were undiagnosed. Among the classical ML algorithms, MLPC achieved the highest F1 score (0.9184), while the Logistic algorithm achieved the highest AUC-PR value (0.9631). In contrast, alternative models provided better F1 scores: 0.9155 for ELM and 0.9989 for FLN. Additionally, the AUC-PR values were also quite high, with 0.9138 for ELM and 0.9568 for FLN. In conclusion, this study demonstrates that alternative algorithms provide robust predictions by simultaneously achieving high F1 and AUC-PR values. Despite the imbalanced nature of the dataset, this study yields quite good results in endometrial cancer prediction using only CA125 values as clinical biochemistry inputs. This study is significant in that it addresses the issue of determining endometrial cancer, which is a disease with a high mortality rate, based on blood values in addition to various tests. Additionally, it is anticipated to contribute to an important area in the literature by providing information on whether it is possible to detect cancer risk by examining blood values.