The magnetic component is responsible for the transmission, storage, filtering and other functions of the magnetic energy, which directly affects the volume, weight, loss and cost of the power converter, so it is particularly important to accurately predict the loss of the magnetic core. In order to solve the problem that core loss cannot be accurately evaluated when using magnetic components, a data-driven core loss modeling method is proposed. Firstly, the decision tree and XGBoost model are used to classify the excitation waveform, and the flux density distribution and waveform characteristics of each material in the test set are drawn. Secondly, XGBoost, support vector machine, gradient lifting regression tree and K-nearest neighbor models were established to predict the core loss of the samples in the test set. Then, a single objective optimization model based on genetic algorithm and particle swarm optimization algorithm and a multi-objective optimization model based on NSGA-II algorithm are proposed to obtain the corresponding conditions such as temperature, frequency and waveform under the optimal objective function. The results show that XGBoost has the best effect on waveform classification and core loss prediction, and the accuracy of core loss prediction in training set and test set are 85.66% and 84.83%, respectively. NSGA-II algorithm has the best performance in the joint optimization of core loss and transmitted magnetic energy.