Commercial biopesticides are primarily based on the bacterium Bacillus Thuringiensis which produces insecticidal proteins. Dynamic models have been proposed to describe the production of these proteins, but their approximation can still be improved. Dynamic hybrid modelling combines kinetic models with data-driven algorithms which could be promising to improve the existing dynamic models. This work presents a dynamic hybrid model using Support Vector Machine (SVM) to predict the specific production of protein and spore by different strains of B. thuringiensis. The dynamic hybrid model was trained and validated with independent datasets of batch fermentations using the production rates of biomass, protein and spores computed with a kinetic model and the strain type as predictors. Additionally, Shapley values were calculated to determine the pertinence of each predictor. The Normalized root mean squared errors (NRMSE) revealed an improvement of the dynamic hybrid model of 12% for proteins and 7% for spores over the dynamic model. Although good, the model could be further improved by training the model with a larger quantity of data.