Machine Learning Based Multi Criteria Decision Analysis for Lymphoma Diagnosis.

Abstract

Lymphoma diagnosis is extremely difficult owing to its complexity as a disease, as it exhibits a high level of non-specific symptoms. This paper aims at developing a machine learning algorithm that will improve the accuracy and efficiency of a lymphoma diagnosis. A virtual database containing 2,000 patients' data was designed, with emphasis on clinical and demographic factors. For equality in experimental conditions, meticulous normalization was undertaken before processing the data. LightGBM algorithm was employed as the principal model, with feature importances being identified as significant clinical factors.

It was found that this model performed better compared to others, with a result of AUC=0.8920 for ROC, predicting with an accuracy of 81.25%. Analysis of feature importance showed that Hemoglobin, C-Reactive Protein (CRP), Lactate Dehydrogenase (LDH) had high importance scores for this model, which was expected as per existing literature. This further proves that this model could work as a decision-making tool for a Lymphoma diagnosis. In further studies, it would be beneficial if this model was tested with empirical data sets from various healthcare settings, further improving its accuracy by taking various crucial variables into account for this purpose.