Abstract
This research addresses the challenges posed by the "curse of dimensionality" in machine learning, particularly in the context of non-image data. As the number of features increases, the complexity and computational costs escalate, making accurate model construction difficult. Two primary strategies are employed to mitigate this issue: dimensionality reduction techniques and methods that restrict feature combinations. This study aims to develop a generic DeepInsight framework that systematically transforms variable configuration, feature extraction, and model building processes. By applying this framework to stock price prediction, we seek to enhance the predictive accuracy and validate the effectiveness of Deep Learning approaches in non-image data analysis.
https://ror.org/0447kww10 
