Diagnosis of breast cancer using neural networks with different logistic regressions
Jannatul Ferdaus · Doria (University of Helsinki) · 2020
Breast cancer is one of the prevalent diseases worldwide. Early diagnosis of breast cancer can save a patient’s life and allow the patient to have an effective treatment. Machine learning has been contributing to the medical sector for decades. This dominant subclass of artificial intelligence is able to detect patterns to provide a medical diagnosis. Many breast cancer research has been done through the medium of machine learning in recent years. This thesis aims to diagnose breast cancer by discerning benign tumors from malignant tumors using several machine learning algorithms. Different logistic regression algorithms were employed with neural networks: shallow logistic regression, shallow softmax regression and deep softmax regression. To observe how the performance improvement techniques influence the accuracy of a classifier, dropout, batch normalization and weight regularization were applied to the deep softmax regression model.