RCSAN residual enhanced channel spatial attention network for stock price forecasting
Data preprocessing
The raw encoder input \(X_{\textrm{en}}\) and decoder input \(X_{\textrm{de}}\) undergo a normalization process before being fed into the model, as outlined below:
To standardize the price-based features (such as Open, Close, High, Low, Volume), the Min-Max scaling technique is applied, which transforms the data into the \([0,1]\) range using the formula:
$$\begin{aligned} {X^{(i)}_{\text {norm}} = \frac{X^{(i)} – \displaystyle \min _{j}(X^{(j)})}{\displaystyle \max _{j}(X^{(j)}) – \displaystyle \min _{j}(X^{(j)})}, \quad i = 1, \dots , N} \end{aligned}$$
(1)
Here, \(X^{(i)}\) is the original value of feature \(i\), and the denominator ensures that all features are rescaled to a common…
