Deep Learning based Early Prediction Scheme for Breast Cancer

Breast cancer is one of the rapid spreading diseases resulting in the death of younger age group of women. Unfortunately, as the detection of cancer is at later stage, the lifetime of the patient is decreased. If the detection is at early stage, then their lifetime could have been improved. Hence, the proposal aims at predicting the presence of breast cancer at early stage through deep learning. To identify suitable model for deep learning, initially machine learning algorithm with Logistic Regression, K Nearest Neighbors, Support Vector Machine (linear), Support Vector Machine, Gaussian, Decision Tree and Random Forest along with ensemble learning algorithms such as Bagged Trees, Subspace discriminant and RUSBoosted Trees are implemented with 30 attributes. Comparison of performance metrices indicates that random forest performs better. Then, feature selection of 14 attributes is attained through heat map. With minimal features, the above set of algorithms is implemented and their corresponding performance indices such as accuracy, misclassification cost, prediction speed, training time, predicted class, true class, positive predict value, sensitivity, specificity, precision, F1 score, Area Under the Curve and Receiver Operating Characteristic Curve are obtained. In this, random forest performs better and in addition, the performance of 14 attributes is almost exhibiting closer performance as that of 30. However, feature selection is mandate and can be eliminated if the algorithm is implemented through deep learning model. The model consists of many hidden layers which performs binary classification on the given dataset to predict whether a person is malignant or benign. The performance indices of the proposed model are validated and the results exhibit its supremacy.