Probabilistic AutoRegressive Neural Networks for Accurate Long-Range Forecasting.
International Conference on Neural Information Processing., 2023
Forecasting complex time series requires models that can handle nonlinearity, non-stationarity, and long-range dependencies. This work proposes PARNN, a hybrid Probabilistic AutoRegressive Neural Network that integrates ARIMA-based error feedback into neural architectures. The model provides reliable uncertainty quantification through prediction intervals and conformal prediction.
<img src="/images/PARNN_Model_Image.png" alt=PARNN_Model_image" style="width: 90%; max-width: 800px;">
Architecture of the PARNN framework.