Z-Score Normalized Features with Maximum Distance Measure Based k-NN Automated Blood Cancer Diagnosis System

Leukemia is a blood-forming cancer disease characterized by the abnormal growth of White Blood Cells. Early detection and treatment of cancer reduce mortality and increase the survival rate. Most of the k-NN approaches used a different combination of statistical and geometrical features of the nucleus and cytoplasm with and without normalization, resulting in an uncertain range of features. Minkowski, Euclidean, City-block, Correlation, and Cosine distance metrics discover the average similarity between features predicts Leukemia with less accuracy. This paper proposes a Z-Score normalized feature set combined with a k-NN maximum distance measure-based automated blood cancer diagnosis system to address this problem. Initially, the features are normalized using the Z-score normalization method, which eliminates the mis-classification outcomes. The Chebyshev distance measure, which finds the maximum similarities between the features when k=1, has the highest accuracy of 97.92%. Furthermore, tuning the parameters by grid-search improves the performance rate by 98.65% .