Feature Spotlight: Unveiling Cancer's Tell-Tale Signs

Finding biomarkers for cancer detection is essential for prompt diagnosis and efficient care. In this project, we used the Breast Cancer Wisconsin dataset to identify possible biomarkers for breast cancer using machine learning techniques. Our primary aim was to implement logistic regression and decision tree classifiers to identify significant features that can predict cancer presence with high accuracy. Initially, we pre-processed the dataset by handling missing values and standardizing the features. Next, we trained and analyzed decision tree and logistic regression models. The decision tree model had an accuracy of 94%, while the logistic regression model had a 98% accuracy rate. Important measures like recall, F1 score, ROC AUC, and precision were computed to give a thorough assessment of the models' performance. We carried out feature selection in order to improve interpretability, and we determined which features were most important in predicting breast cancer. The decision tree model's feature importance scores highlighted significant biomarkers, including mean radius, mean texture, and mean perimeter. Additionally, we visualized these findings using a correlation matrix and a feature importance bar plot, aiding in the clear interpretation of the results.