Data augmentation and mixup techniques for stock returns prediction
In financial forecasting, the caliber of data underpinning models is critical for precision in predicting market trends. This research investigates the efficacy of data aug- mentation, a technique widely used in domains such as image processing, applied to financial time series. We assess how method...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2024
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/175462 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | In financial forecasting, the caliber of data underpinning models is critical for precision in predicting market trends. This research investigates the efficacy of data aug- mentation, a technique widely used in domains such as image processing, applied to financial time series. We assess how methods like Cut Mix, Linear Mix, and Amplitude Mix—augmented with filters such as Savgol, Kalman, and LOWESS—can refine predictive models. The hypothesis posits that these augmentations can uphold data integrity while infusing beneficial variability to enhance model robustness and forecasting accuracy. Results confirm that Linear Mix and Amplitude Mix improve classification performance by introducing manageable variability without sacrificing the data’s core structure. In contrast, Cut Mix often decreases accuracy, underscoring the potential drawbacks of excessive data modification. The findings suggest a strategic equilibrium in data augmentation is paramount for advancing the adaptability and accuracy of financial forecasting tools. |
|---|