This paper presents a deep learning model for predicting traffic flows, focusing on capturing sharp nonlinearities due to transitions between free flow, breakdown, recovery, and congestion. The model combines a linear model fitted with $\ell_1$ regularization and a sequence of tanh layers to identify spatio-temporal relations among predictors. The methodology is applied to road sensor data from Interstate I-55, predicting traffic flows during two special events: a Chicago Bears football game and an extreme snowstorm. The results demonstrate that deep learning provides precise short-term traffic flow predictions, even in scenarios with sudden regime changes. The paper also discusses the use of variable selection techniques, such as hierarchical sparse vector autoregressive models and dropout, to improve model performance. The deep learning model is shown to outperform traditional methods in capturing nonlinear spatio-temporal effects and providing accurate forecasts for both recurrent and non-recurrent traffic conditions.This paper presents a deep learning model for predicting traffic flows, focusing on capturing sharp nonlinearities due to transitions between free flow, breakdown, recovery, and congestion. The model combines a linear model fitted with $\ell_1$ regularization and a sequence of tanh layers to identify spatio-temporal relations among predictors. The methodology is applied to road sensor data from Interstate I-55, predicting traffic flows during two special events: a Chicago Bears football game and an extreme snowstorm. The results demonstrate that deep learning provides precise short-term traffic flow predictions, even in scenarios with sudden regime changes. The paper also discusses the use of variable selection techniques, such as hierarchical sparse vector autoregressive models and dropout, to improve model performance. The deep learning model is shown to outperform traditional methods in capturing nonlinear spatio-temporal effects and providing accurate forecasts for both recurrent and non-recurrent traffic conditions.