A predictive model that forecasts future returns within a universe of stocks has been presented. The model uses data science techniques to arrive at the forecast and operates without any sector constraints. The ranking is based on forecasted future excess returns. The approach includes variable selection (random forest), nonlinear interaction detection( gradian boosting and fractional polynomials), regularization and validation using rolling window method. The predictive power of the proposed methodology is illustrated through directional accuracy comparison between top 15% and down 15% of stocks.