This paper introduces a new semi-parametric methodology for the implied volatility surface, which incorporates machine learning algorithms. Given a starting model, a tree boosting algorithm sequentially minimizes the residuals of observed and estimated implied volatility. To overcome the poor predicting power of existing models, we include a grid in the region of interest, and implement a cross-validation strategy to find an optimal stopping value for the tree boosting. Back testing the out-of-sample performance on a large data set of implied volatilities on S&P 500 options, we provide empirical evidence of the strong predictive potential of our methodology.