Creating New Variables Using Deep Learning for Training Classical Machine Learning Models
Creating New Variables Using Deep Learning
for Training Classical Machine Learning Models
Arakelyan Garnik
Summary
Key words: neural networks, logistic regression, neural variables, scoring model, ROC curve, PR curve, qualitative indicators
This study explores the use of neural networks to generate new variables – referred to as neuro-features – that help reveal hidden relationships in the data. These features can then be used in classical machine learning algorithms to improve their performance. As a case study, the use of neuro-features in logistic regression is examined. To assess the impact of these features, two logistic regression models were built: one without neuro-features and one with them. The following performance metrics were calculated for each model: Accuracy, Precision, Recall, F1 Score, ROC AUC, and PR AUC. A t-test was also conducted to evaluate the statistical significance of the performance difference. The results of the study show that incorporating neuro-variables into existing datasets during model development can significantly improve model performance metrics and contribute to the model’s stability. The practical significance of the findings lies in the fact that neural networks can be utilized even with limited data availability. In such cases, new variables can be generated to enhance the effectiveness of other machine learning models. This opens the door for applying machine learning technologies even in data-scarce environments, where traditional methods may be less effective.