Classification of Ovarian Cancer by Machine Learning Models and its Analysis by Explainable Artificial Intelligence

Ovarian cancer is a cancerous tumor in the ovary. To assist with its diagnosis, machine learning models can be used to predict ovarian cancer based on the data available from medical tests. We use data (in terms of features for a number of patients) from Kaggle and according to the mutual information scores of the features, human epididymis protein 4 and carbohydrate antigen 125 appear to be the most important features in the dataset. We came up with a framework that employs a variety of machine learning models and a neural network to predict whether a patient has ovarian cancer or not. Most of the machine learning models are fairly good for prediction, however, our findings indicate that Random Forest, with accuracy of 85.4 %, is the best possible classifier; it exhibits a recall of 85 % to demonstrate its effectiveness in having less false negatives. Neural network like Multi-layer Perceptron also has the ability to perform almost as good as Random Forest if we use a dataset containing some of the best features. Further, we explain the prediction of a test instance by means of an explainable artificial intelligence tool called Local Interpretable Model Agnostic Explanations (LIME).