 We have developed a novel framework for handling missing values in time series data. The framework uses three different missing value fixing algorithms, which are then incorporated into a deep recurrent neural network, DRNN. This DRNN is composed of long short-term memory, LSTM, layers and fully connected layers, and is trained with real, world air quality and meteorological data sets from Jingjingi area, China. We compared this framework with two other baselines, a deep feed forward neural network, DFNN, and a gradient boosting decision tree, GBDT. Experimental results showed that our proposed DRNN framework outperformed both DFNN and GBDT, demonstrating its effectiveness in handling missing values in time series data. This article was authored by Jfan, Culey, and others.