Several bodies of work have demonstrated a correlation between a stellar host’s chemical composition and the presence of planetary companions, particularly in the case of gas-giant planet occurrence. Assuming that there is a generalisable transformation function which can generate a probability score of a star being a gas-giant host, stellar chemical abundance datasets could be used to train an ML classifier capable of discriminating between the host and comparison samples. Since abundance data tends to be incomplete, imputation techniques are also required before training. To reach both goals simultaneously, that is, a generalisable classification model and reliable imputation, we employ a multi-objective optimisation model based on the NSGA-II algorithm. This allows for an effective search of the parameter space for Pareto solutions of the ideal combination of hyperparameters for both objectives. Within the algorithm, we use a fuzzy clustering model for the feature imputation and several ML algorithms for classification. Preliminary results show optimisation of both fitness functions, with reliable generalisation during cross-validation.