A weighted-likelihood framework for class imbalance in Bayesian prediction models