Volatile organic compounds (VOCs) are continuous medical data regularly studied to perform non-invasive diagnosis of diseases using machine learning tasks for example. The project PATHACOV aims to use VOCs in order to predict invasive diseases such as lung cancer. In this context, we propose to use a multi-objective modeling for the partial supervised classification problem and the MOCA-I algorithm specifically designed to solve these problems for discrete data, to perform the prediction. In this paper, we apply various discretization techniques on VOCs data, and we analyze their impact on the performance results of MOCA-I. The experiments show that the discretization of the VOCs strongly impacts the classification task and has to be carefully chosen according to the evaluation criterion.