High-throughput catalytic reactors are useful for developing large datasets of kinetic results and discovering faster new catalysts and models. However, these large datasets are often challenging to curate, and here we present a methodology for reinforced outlier detection using a robustified principal component analysis with a minimum covariance determinant. We estimate the kinetic parameters of three Mn-Na-W catalysts for the oxidative coupling of methane using datasets curated in different ways. These catalysts differ in their support composition (SiO2 and SiC), preparation method (impregnation or spray-dying), and performance. The kinetic model subsequently trained consists of six global reactions, and the obtained fit demonstrates that the proposed data curation method is a useful tool for enhancing the predictions of subsequent models and gaining further insight into underlying correlations within the data, especially when the aesthetic curation of data is too resource-demanding and/or biased.