Machine learning based analytical system for predictive detection of Leukemia using WEKA

Abstract

In recent times, classification of leukemic blood cell by using machine learning techniques has gained the attention of many researchers for developing an automated model which can assists doctors in detection of leukemia. Also, it is quite challenging to accurately predict the blood cancer as symptoms are very general in initial stages. In this manuscript, we have presented an approach for predictive detection of leukemia by observing the important features from the blood test and using various classifiers. We have observed that AdaBoostM1 classification algorithms gives better result than Bayes Net classifier. We have also derived some most important features age, infected (“Yes”, ”No”), white blood cell count, red blood cell count, platelet count, leucocytes count, mch, hemoglobin, hematocrit, neutrophils, eosinophils, lymphocytes, monocytes, basophils, mpv, nrbc hash, diastolic blood pressure, total cholesterol, triglycerides, hdl cholesterol, which has significant impact on leukemia detection. We have achieved 98.50% accuracy, 96.99% sensitivity, 98.7% specificity and 98.30% precision values for detection of leukemia by using Random Forest classifier.