Output details

Partial logistic artificial neural network for competing risks regularized with automatic relevance determination.

<24> Machine learning approaches. The significance of this paper is to define a rigorous machine learning model of survival taking account of competing risks. This was achieved by extending a previously defined Bayesian regularisation methodology applied to single-risks, taking into account the theoretical constraints which arise from the interactions between the different risks, since observation of one event automatically censors the other risks. The originality of the method is the rigorous application to the partial logistic artificial neural network of the theoretical foundations of competing risks methodologies. The method was validated on one of the world’s largest breast cancer databases.