Methods of Machine Learning and Data Mining for Analyzing Medical Data

There are several methods for treatment of Acute Lymphoblastic Leukaemia (ALL), having the same effectiveness in average. Each of those treatments helps a specific subgroup of patients, while it can be not so effective for another one. This fact leads to the need of developing an individualized treatment rule that would recommend optimal treatment to a patient according to his/her features. In this work we consider a machine learning approach to generate such a rule with randomized clinical trial data as training sample. To obtain the optimal rule, we compare prediction accuracy of several machine learning methods such as an estimation based on L1-penalised least squares, random forest and lasso-regression. The method with the highest accuracy in a risk group is used for optimal treatment recommendation in this risk group.