An Effective Leukemia Prediction Technique Using Supervised Machine Learning Classification Algorithm

Abstract

Leukemia is not only fatal in nature, the treatment is also extremely expensive. Leukemia’s second stage (typically there are four stages) is enough to blow a large hole in a family’s savings. In this paper, we have designed a supervised machine learning model that accurately predicts the possibility of Leukemia at an early stage. We mainly focus on regular symptoms and the probabilities of a subject to develop Leukemia later on. The parameters or features are usually information available at regular checkups. Firstly, we have defined 17 parameters in consultation with the specialist doctors and then we have collected primary data through surveys of different Leukemia and Non Leukemia patients from hospitals. We have divided the data into train and test datasets and applied different machine learning algorithms such as Decision Tree, Random Forest, KNN, Linear Regression, Adaboost, Naive Bayesian, etc. to find out the accuracy. We obtained 98% of accuracy using Decision Tree and Random Forest, 97.21% using KNN, 91.24% using Logistic Regression, 94.24% using Adaboost, and 75.03% using Naive Bayesian, respectively. It is observed that the Decision Tree and the Random Forest classifier outperform the rest.

Supported by Institute of Advance research of United International University (UIU) Research Project No. UIU/IAR/02/2019-20/SE/07 We sincerely acknowledge Dhaka Shishu (Children) Hospital and National Institute of Cancer Research & Hospital (NICRH) for providing us training and blood sample data.