<p>Composite materials with exceptional insulation properties hold immense appeal for advanced electrical and electronic devices. However, owing to extensive design parameters, the insulation properties are intricately related to constituent elements and structural configurations. Traditional design methods generally adopt exhaustive trial-and-error experiments, requiring costly and labor-intensive endeavors for optimization. The electrical tree simulated by finite element analysis (FEA) is a key indicator of insulation properties, with a longer developing time indicating enhanced insulation performance. It is capable of offering precise simulations, yet does so at a substantial computational expense. Herein, we present the accurate and efficient prediction of electrical tree developing time in dielectric composites using a machine learning (ML) model. The ML model is well-trained utilizing the gradient-boosted decision tree (GBDT) algorithm, drawing upon a database established with FEA models. The GBDT model exhibits remarkably low prediction errors, serving as an effective alternative to the FEA model that significantly boosts the prediction efficiency. As a proof of concept, the GBDT model is applied to predict existing experimental results, further validating its feasibility and reliability. Besides, the contribution of each input feature to prediction results is analyzed, offering invaluable insights that guide the design of dielectric composites.</p>