<p>Rubber compound development usually caters to a particular engineering application, which can require a combination of properties. The process starts with choosing a starting point formulation, which yields properties closer to the required properties (target). To achieve this target, the formulation must be changed and tested for its properties. The trial-and-error method is an old technique for compound design that typically requires significant time, energy, and resources, as well as technical experts with extensive knowledge and experience. The major requirement is the effective modelling of rubber and ingredients with their properties to reduce cost and improve sustainability and efficiency. This study utilised and explored various data-driven approaches for compound design by addressing complex relationships using artificial intelligence models and machine learning algorithms, without requiring extensive laboratory experiments. The artificial neural networks, random forest, decision tree, and extreme gradient boosting (XGBoost) regression models were trained on natural rubber-based formulation data and presented alongside three corresponding mechanical properties. The compound development process predicted and optimiSed rubber formulations, properties, and ingredient levels for specific applications using the database. The outstanding performance of the XGBoost model, which achieved an R² value of up to 0.96, demonstrates the advancements in rubber formulation design through machine learning. The study uncovered encouraging results as well as potential challenges for the future application of machine learning in rubber compound design.</p> Graphical abstract <p></p>