The Research About Breast Cancer Prediction Model

Abstract

Breast cancer poses a major threat to the health of women worldwide. This study analyzed data from breast cancer patients to develop a predictive model for identifying cases of the disease that are malignant based on cellular measurements. The dataset from the University of California, included 569 instances of 10 numerical variables such as radius, texture, and concavity of the cell samples. After initially exploring the relationships between the variables, the study used methods such as logistic regression and model training. Radius, perimeter and area were integrated since they are positively correlated. Concavity represents concave points, as both describe depressions in the cell outline. Fractal dimension and compactness were combined into a new predictor, F/C. Logistic regression analysis revealed that radius and concavity had the highest prediction accuracies of 87.9% and 88.1%, respectively. Compactness performed moderately well, while the fractal dimension had little diagnostic value. The accuracy of the F/C variable improved by 85.1% over compactness alone. A multi-variable model combining radius, concavity and F/C further improved accuracy and specificity to 92.1%. However, no single variable perfectly predicted cancer diagnosis, suggesting that the data patterns were complex. Further interactions between variables could be uncovered by advanced modelling. In conclusion, the study suggests that composite measures such as radius and concavity are better predictors of breast cancer than isolated factors. More comprehensive clinical data and sophisticated analytic techniques need to be built up to improve diagnostic performance. The model sets the stage for improving breast cancer prognosis through data-driven prediction.