Understanding and predicting fruit phenotypes during development is crucial for quality improvement and food industry applications. Metabolomics, which analyzes the complete set of metabolites within biological samples, is particularly interesting in the case of multi-species studies. It offers a global view of the biochemical processes underlying phenotypes and provides data for many metabolites shared between species, which are therefore interoperable variables. In this study, we combined metabolomics data with machine learning techniques to predict diverse phenotypic traits across the development of ten fruits. By integrating metabolomics profiles with phenotype annotations, we constructed predictive models capable of associating metabolic variable abundance with traits such as growth rate, developmental stage, acidity or sugar content all along fruit development. Supervised machine learning algorithms, including Ridge, Elastic-Net and LASSO regression, Random Forests and Support Vector Machines were used to capture the complex relationships between metabolic profiles and phenotypic variations. Feature selection methods were used to identify key metabolic variables driving the prediction of each phenotype, providing insights into the metabolic functions potentially governing fruit development. Cross-validation procedures and independent validation datasets were employed to assess the robustness and generalization performance of the predictive models. In conclusion, the application of metabolomics on this multispecies datasets represents a significant advancement in our understanding of fruit development and offers unprecedented opportunities for innovation. By combining the power of metabolomics and advanced machine learning techniques, we will be able to unravel intricate molecular mechanisms governing phenotypic traits across multispecies experiments.