Pastprop-RNN: improved predictions of the future by correcting the past

2021/06/25 by André Baptista, A. Baptista, Yassine Baghoussi +9
Computer Science · Decision Sciences · #Anomaly (physics) #Artificial intelligence #Artificial neural network #Backpropagation #Baseline (sea) #Benchmark (surveying) #Computer science #Data mining #FOS: Computer and information sciences #Forecasting Techniques and Applications #Machine Learning (cs.LG) #Machine learning #Stock Market Forecasting Methods #Time Series Analysis and Forecasting #cs.LG

paper · pdf · doi:10.48550/arxiv.2106.13881

published in arXiv (Cornell University) (Cornell University)

arxiv created 2021/06/25 · openalex publication_date 2021/06/25 · arxiv updated 2021/06/29 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Forecasting accuracy is reliant on the quality of available past data. Data disruptions can adversely affect the quality of the generated model (e.g. unexpected events such as out-of-stock products when forecasting demand). We address this problem by pastcasting: predicting how data should have been in the past to explain the future better. We propose Pastprop-LSTM, a data-centric backpropagation algorithm that assigns part of the responsibility for errors to the training data and changes it accordingly. We test three variants of Pastprop-LSTM on forecasting competition datasets, M4 and M5, plus the Numenta Anomaly Benchmark. Empirical evaluation indicates that the proposed method can improve forecasting accuracy, especially when the prediction errors of standard LSTM are high. It also demonstrates the potential of the algorithm on datasets containing anomalies.

Citations

• Recurrent Neural Network Regularization
• On the difficulty of training Recurrent Neural Networks
• Long Short-Term Memory
• The Exponentially Weighted Moving Average
• Doctor AI: Predicting Clinical Events via Recurrent Neural Networks
• Evaluating Real-time Anomaly Detection Algorithms - the Numenta Anomaly Benchmark
• Deep Learning is Robust to Massive Label Noise
• Machine Learning vs Statistical Methods for Time Series Forecasting: Size Matters
• A Comparative Analysis of Forecasting Financial Time Series Using ARIMA, LSTM, and BiLSTM
• A Data Quality-Driven View of MLOps

Related

• Predicting the Future by Retrieving the Past
• IdealTSF: Can Non-Ideal Data Contribute to Enhancing the Performance of Time Series Forecasting Models?
• Time Series Forecasting Models Copy the Past: How to Mitigate
• Forecasting with Deep Learning: Beyond Average of Average of Average Performance
• Abstain Mask Retain Core: Time Series Prediction by Adaptive Masking Loss with Representation Consistency