Advanced machine learning algorithms like Artificial Neural Networks(ANNs) can’t model time-dependent data without some pre-processing. The additional processing hurdle often deters forecasters from implementing advanced methods in favor of classic (but less powerful) approaches. However, I’ve observed some notable accuracy gains applying ANNs to forecasting problems. Accordingly, this post provides a basic playbook for data cleaning, feature engineering, model selection, prediction, and risk assessment when forecasting with Neural Nets.