Estimation of predictive performance in high-dimensional data settings using learning curves