SeqOAE: Deep sequence-to-sequence orthogonal auto-encoder for time-series forecasting under variable population sizes

Abdallah A. Chehade, Wael Hassanieh, Vasiliy V. Krivtsov

Abstract: Highlights•Addresses data maturation phenomenon by mapping immature to mature observations.•Supports time-series forecasting under variable population sizes.•Proposes a non-parametric, sequence-to-sequence, nonlinear and robust deep learning model.•Leverages orthogonal deep latent features for long-term forecasting of mature time-series data.•SeqOAE outperforms existing models in warranty claims forecasting.