The challenges are learning from data with missing values and finding shared representations for multimodal data to improve inference and prediction. The proposed method uses deep learning technique to reduce the volume of missing data which occur due to low battery and transmission loss. In order to find missing data imputation and new modality prediction, the original incomplete raw data is trained and tested using the stacked autoencoder to predict the corresponding missing value. Deep Multimodal Encoding methods use intra and intermodal learning to compute a new modality prediction. Deep Multimodal Encoding (DME) can achieve a Root Mean Square Error (RMSE) of the missing data imputation which is only 20% of the traditional methods. The performance of Deep Multimodal Encoding is robust to the existence of missing data.