Early cancer predictions using ensembles of machine learning and deep learning
Wasiur Rhmann, Babita Pandey · 2021
Cancer is the most frequent cause that is responsible for large number of deaths globally. According to the report published in 2018 by the international agency of research on cancer, one man in five and one women in six in the whole world develops cancer during their life. One in eight men and one in eleven women in the whole world die of cancer each year. Lung cancer and breast cancer are two major types of cancer with the highest numbers of new cases. Breast cancer is a prevalent and the second most deadly disease among women after lung cancers that are responsible for deaths. Survival of cancer patients largely depends on timely and accurate diagnosis of disease. In literature, various machine learning (ML) techniques are applied for breast cancer prediction. ML techniques utilize past data for the prediction of disease. Identification of tumors as benign and malignant is a crucial part in the detection of cancer and is considered as a classification problem. In this chapter, ensemble- based techniques and deep learning (DL) techniques are used for lung and breast cancer prediction. Two datasets, Wisconsin and Coimbra, obtained from wellknown UCI ML repository are used for experimental purposes. Ensemble techniques are used to make effective classifiers with improved prediction capability. Generally, ML classifiers like Logistic regression (LR), random forest (RF), naïve Bayes (NB), and support vector machine are used for prediction. Combinations of different classifiers are used to enhance the performance of single classifier, and combined classifier is known as an ensemble. In recent years, a subfield of ML, DL, has emerged as a promising area with several techniques that have drastically boosted the performances of models that have caught the attention of researchers. DL originated from neural network. Although DL techniques have shown very good performance on different types of problems, they are computationally intensive. In this chapter, ensemble ML techniques have shown best performance on two cancer datasets, Coimbra and lung cancer datasets, while deep neural network has shown best results for Wisconsin dataset, and ensemble technique is very close to it.