Evaluating the Impact of Hybrid Deep Learning Model and Data Balancing Techniques on Classification Performance: A VEREMI Dataset Analysis

Abstract

The VEREMI dataset is used in this study to look into how well mixed deep learning models and data balancing methods work for binary and multiclass classification. The study uses many preprocessing steps, such as label encoding, data scaling, and handling missing values. It also uses the SMOTE method to fix data mismatch. Three models Autoencoder, Long Short-Term Memory (LSTM), and VEREMI_LA, a new hybrid model that combines Autoencoder and LSTM were learned and tested on datasets that were both balanced and skewed. In binary classification, the results show that the mixed VEREMI_LA model does better than individual models. It achieves an amazing 99.9% accuracy in both balanced and skewed situations. The F1-score also shows that the VEREMI_LA model has better performance; it consistently hits 99.9%, showing that it can handle different types of data. With an accuracy of only 50% and an F1-score of 34%, the Autoencoder model, on the other hand, did much worse, especially on the balanced dataset. In the multiclass classification task, the VEREMI_LA model once again proved to be the best, getting 86% of the correct answers on the uneven dataset and 97% of the correct answers on the balanced dataset. With only 52% accuracy, the Autoencoder, on the other hand, did the worst on the balanced sample. These results show that the mixed method works to make classification more accurate and reliable. The study stresses how important data balance methods and combining mixed models are for improving the performance of machine learning for difficult classification tasks. This study gives useful information to people who want to make classification models work better in the real world. It suggests that mixed methods like VEREMI_LA can greatly improve the accuracy of predictions in a wide range of datasets.