Issue with Wavelet Denoising in Stock Price Prediction

[vanthomiy]

The paper describes the wavelet denoising process as follows:

The wavelet denoising process played a crucial role in enhancing model accuracy. By transforming noisy data into a cleaner form, the models could achieve higher prediction accuracy, as demonstrated by the improved F1 scores and other performance metrics.

The denoising process adjusts data using 100 values by removing outliers and aligning data points. However, this alignment is not constrained to a time-backwards-only approach, causing future data points to influence current values. This introduces data leakage, where future information is exposed during both training and testing.

The model performs well when tested with denoised data, as this data already incorporates future values. However, when tested with raw, noisy data, the results drop to random guessing. This reliance on denoised data makes real-world prediction impossible, as future data would not be available during actual deployment.