Breast Cancer Detection Using Advanced Data Mining and Machine Learning Techniques

Abstract

Breast cancer is one of the world's most serious health issues; it is the most commonly diagnosed cancer in women, and prevention appears impossible because the cause is unknown; as a result, early detection is critical to the patient's prognosis. In developing nations such as Iraq, where access to specialised healthcare is limited, radiologists are in short supply and routine clinical check-ups are rare. In the current work, the algorithms used in data mining for early breast cancer diagnosis were applied to the data set obtained from the UCI database (Original Wisconsin Breast Cancer Database). Medical experts obtained this data set as a result of sensitive and laborious work that lasted 4 years between 2016 and 2020. Data mining algorithms were applied to the data set with the WEKA data mining program, and the results produced by the algorithms were compared. The developed methods were analysed, and accuracy rates were calculated. Tests were also carried out on the running times of the algorithms used. Results were obtained by setting the "Percentage Split" percentage separation section to 66% in the test settings section of the WEKA program. All processes were performed with the same settings and 10-fold cross-validation. The results are as follows: decision trees 72%, random forest 75%, k-nearest neighbours 66%, naive Bayes 60%, and logistic regression. It was discovered to be 73%, with multilayer artificial neural networks accounting for 66%. It is evident from the results that cross-validation success values are extremely low. Furthermore, in cross-validation, a newly added subject's test data may be either test or training data; however, in the percentage separation system, it will be directly associated with the test portion. Finally, it was concluded that it will also help medical professionals make faster and healthier decisions in the early diagnosis of breast cancer.