Differential Machine Learning for Time Series Prediction
This work addresses the problem of improving time series prediction accuracy for applications in fields like finance and chaos theory, offering a novel method that is incremental in enhancing existing neural network architectures.
The paper tackles the challenge of accurate time series prediction by proposing a differential learning approach that trains neural networks on both original and differential series, resulting in a Diff-LSTM network that outperforms prevalent models like RNNs, CNNs, and other LSTM variants on chaotic and real-world financial datasets.
Accurate time series prediction is challenging due to the inherent nonlinearity and sensitivity to initial conditions. We propose a novel approach that enhances neural network predictions through differential learning, which involves training models on both the original time series and its differential series. Specifically, we develop a differential long short-term memory (Diff-LSTM) network that uses a shared LSTM cell to simultaneously process both data streams, effectively capturing intrinsic patterns and temporal dynamics. Evaluated on the Mackey-Glass, Lorenz, and Rössler chaotic time series, as well as a real-world financial dataset from ACI Worldwide Inc., our results demonstrate that the Diff- LSTM network outperforms prevalent models such as recurrent neural networks, convolutional neural networks, and bidirectional and encoder-decoder LSTM networks in both short-term and long-term predictions. This framework offers a promising solution for enhancing time series prediction, even when comprehensive knowledge of the underlying dynamics of the time series is not fully available.