Data Analytics And Computational Modelling On Breast Cancer Data

Breast cancer has become the most common female cancer around the world.Early diagnosis and treatment are important for declining the incidence and mortality rate of breast cancer.Computer-aided diagnosis system can improve diagnostic accuracy.While there is a lack of research on analysing the relationship between different features and those features'contribution to breast cancer diagnose.In addition.,most of machine learning algorithms need to manually set parameters,but it is hard to set a proper parameter.In order to solve those problems abovementioned,the thesis focused on doing the following researches:(1)Using Bayesian network to quantitatively analyse the relationships between different breast cancer features based on ultrasonic test and fine needle aspiration cytology(FNAC)test.In the same time,features5 contribution to breast cancer diagnosis was analysed.The results showed that the strength of influence between features was different.Features'diagnosis value also was different:in ultrasonic test,SHAP was the most valuable feature;for FNAC test,bare nuclear was the most important feature.(2)Employing various novelty detection technologies to diagnose breast cancer,and test the performance of level set methods using in novelty detection,proposed a proposal using level set methods to diagnosis breast cancer.The results illustrated K-means algorithm was the best detector in ultrasonic test,while for FNAC data K-nearest neighbour algorithm had the highest performance.Besides,the result also showed level set methods can describe data very well.(3)Handling the missing data in the two breast cancer dataset with different methods.The result showed that missing data in ultrasonic feature should use different methods deal with;random forest algorithm was the best method for handling the missing data in FNAC data.