Prediction Approach of Breast Cancer using Dimensionality Reduction and Outlier Detection
Ismot Rimi, Md. Nazrul Islam Mondal, Jakia Oishy · 2022
Breast cancer has become an alarming issue throughout the world. Like other cancers, in breast cancer, the cell loses the ability to keep dividing in a controlled way and produces more cells similar to it and form tumors. Machine learning has a huge impact on prediction of diseases accurately. However, accuracy in predicting the disease also depends on its dataset. Our research work is to remove the outliers and also reduce the dimensionality of the dataset which will increase the accuracy, recall, and precision rate compared to existing models. In our study, we examine the impact of outliers and dimensionality reduction on the Wisconsin Diagnostic Breast Cancer Dataset, which is tested using three different machine learning classifiers. The results show that after the removal of outlier, the proposed technique is applied for dimensionality reduction using kernel-PCA then linear regression, Naive Bayes Classifier and decision tree classifier are applied and the highest accuracy are respectively (kernel cosine-logistic regression) 96.03%, (ker-nel cosine-Naive Bayes) 94.44%,(kernel sigmoid-decision tree) 93.65%. The two suggested strategies outperforms the unfiltered data in terms of accuracy. As a result, the efficiency of the breast cancer prediction is increased. We propose an approach where outliers are checked first for every features so that the feature with maximum outliers can be detected and outliers are removed. Then the dimensionality reduction is done.