LDSVM: Leukemia Cancer Classification Using Machine Learning

Leukemia is blood cancer, including bone marrow and lymphatic tissues, typically involving white blood cells. Leukemia produces an abnormal amount of white blood cells compared to normal blood. Deoxyribonucleic acid (DNA) microarrays provide reliable medical diagnostic services to help more patients find the proposed treatment for infections. DNA microarrays are also known as biochips that consist of microscopic DNA spots attached to a solid glass surface. Currently, it is difficult to classify cancers using microarray data. Nearly many data mining techniques have failed because of the small sample size, which has become more critical for organizations. However, they are not highly effective in improving results and are frequently employed by doctors for cancer diagnosis. This study proposes a novel method using machine learning algorithms based on microarrays of leukemia GSE9476 cells. The main aim was to predict the initial leukemia disease. Machine learning algorithms such as decision tree (DT), naive bayes (NB), random forest (RF), gradient boosting machine (GBM), linear regression (LinR), support vector machine (SVM), and novel approach based on the combination of Logistic Regression (LR), DT and SVM named as ensemble LDSVM model. The k-fold cross-validation and grid search optimization methods were used with the LDSVM model to classify leukemia in patients and comparatively analyze their impacts. The proposed approach evaluated better accuracy, precision, recall, and f1 scores than the other algorithms. Furthermore, the results were relatively assessed, which showed LDSVM performance. This study aims to successfully predict leukemia in patients and enhance prediction accuracy in minimum time. Moreover, a Synthetic minority oversampling technique (SMOTE) and Principal compenent analysis (PCA) approaches were implemented. This makes the records generalized and evaluates the outcomes well. PCA reduces the feature count without losing any information and deals with class imbalanced datasets, as well as faster model execution along with less computation cost. In this study, a novel process was used to reduce the column results to develop a faster and more rapid experiment execution.