An Efficient Approach to Improve Diagnostic Accuracy into Findings of Breast Cancerous Cells with SVM and Logistic Regression Models

This paper consists of an in-depth analysis of two popular supervised learning techniques: Support Vector Machine (SVM) classifier and Logistic Regression. This work aims to demonstrate the working principle and effectiveness of these two algorithms using the breast cancer wisconsin (Diagnostic) dataset extracted from the UC Irvine machine learning repository and predict the malignant and benign characteristics of the same. The dataset was split into a training and testing dataset by selecting 60, 70 and 80 percent of all samples in the dataset respectively as the training dataset. The execution was deployed in ‘google colab notebook’, supporting Python 3.6 programming language. SVM provided an accuracy of 95.90% while logistic regression provided an accuracy as 97.66%. Different quality performance metric parameters such as sensitivity, specificity, positive and negative likelihood ratios, etc. were also obtained for the 70% training dataset level. A 2D scatter plot was also given using the BreakHis dataset trained with the Resnet-50 model.