Abstract:This work addresses the problem of analyzing multi-channel time series data %. In this paper, we by proposing an unsupervised fusion framework based on %the recently proposed convolutional transform learning. Each channel is processed by a separate 1D convolutional transform; the output of all the channels are fused by a fully connected layer of transform learning. The training procedure takes advantage of the proximal interpretation of activation functions. We apply the developed framework to multi-channel financial data for stock forecasting and trading. We compare our proposed formulation with benchmark deep time series analysis networks. The results show that our method yields considerably better results than those compared against.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is multi - channel time - series data analysis, especially its application in stock prediction and trading. Specifically, the paper proposes an unsupervised fusion framework named ConFuse based on Convolutional Transform Learning (CTL) for processing multi - channel data.
### Main Problem Description
1. **Multi - channel Time - series Data Analysis**:
- Many practical problems can be represented as multi - channel time - series analysis, such as demand forecasting, power consumption estimation, biomedical signal processing, etc.
- In these application scenarios, the data of each channel needs to be effectively processed and fused to obtain better prediction or classification results.
2. **Limitations of Existing Methods**:
- Traditional deep - learning methods such as LSTM and GRU are theoretically suitable for time - series data, but they have high hardware requirements, are time - consuming, and are difficult to handle very long sequences.
- Other methods, such as converting time - series into matrix form and using 2D CNN for analysis, although perform well in some fields, still have deficiencies in multi - channel data processing.
3. **The Need for Unsupervised Learning**:
- Most existing multi - channel data analysis methods rely on the supervised end - to - end machine - learning paradigm, which limits the flexibility of the model.
- Unsupervised learning methods can provide greater flexibility and can avoid the complexity of training multiple models for different tasks.
### ConFuse Solution
- **Convolutional Transform Learning (CTL)**: Generate feature vectors by learning a set of convolutional filters to represent multi - channel time - series data.
- **Fusion Mechanism**: After the data of each channel undergoes an independent 1D convolutional transformation, it is fused through a fully - connected layer.
- **End - to - End Training**: Using the approximate interpretation of activation functions, an efficient stochastic gradient descent training strategy is proposed to ensure the effective learning of model parameters.
### Application Scenarios
- **Stock Prediction**: Predict the opening price, closing price, highest price, lowest price, and net asset value of the next trading day.
- **Stock Trading**: Decide whether to buy or sell stocks.
### Experimental Results
- **Stock Prediction**: ConFuse significantly outperforms existing methods in terms of Mean Squared Error (MAE), especially when using SELU and ReLU activation functions.
- **Stock Trading**: ConFuse also performs well in terms of precision, recall, F1 - score, AUC, and Annualized Return (AR), especially when using SELU and ReLU activation functions.
### Summary
By introducing unsupervised convolutional transform learning and an effective fusion mechanism, ConFuse solves the key problems in multi - channel time - series data analysis and achieves results significantly better than existing methods in stock prediction and trading tasks.