Hybrid Multi-objective Machine Learning Classification in Liver Transplantation

This paper constructs a hybrid, multi-objective and evolutionary algorithm based on Differential Evolutions using neural network models and q-Gaussian basis units in order to develop an efficient and complete system for donor-recipient assignment in liver transplantation. The algorithm is used for the classification of a binary dataset and will predict graft survival at 15 and 90 days after the transplantation. Other hybrid approaches combining artificial neural networks with evolutionary computation and well-known algorithms are presented in order to compare the obtained performance of both mono and multi-objective methods, using other methods such as Support Vector Machines and Discriminant Analysis. Some supervised attribute selection methods were previously applied, in order to extract the most discriminant variables in the problem presented. The models obtained allowed medical experts to predict survival rates and to come to a fair decision based on the principles of justice, efficiency and equity.