A new hybrid model for improving outlier detection using combined autoencoder and variational autoencoder.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• High-dimensional datasets present significant challenges for accurate outlier detection.
• Existing methods like Autoencoders (AE) and Variational Autoencoders (VAE) have limitations in stability and performance.

Purpose of the Study:

• To introduce a novel hybrid model, AVE (Autoencoder-Variational Autoencoder), for enhanced outlier detection.
• To evaluate the performance of the AVE model on diverse, high-dimensional datasets.

Main Methods:

• Developed a hybrid architecture integrating AE's reconstruction capabilities with VAE's regularized latent space.
• Conducted extensive experimental evaluations on 16 standard benchmark datasets from various domains.

Main Results:

• The AVE model demonstrated superior performance compared to standalone AE, VAE, and other algorithms.
• Achieved an average precision of 0.6925 and ROC-AUC of 0.8902, significantly outperforming existing methods.
• Secured the best accuracy on 12 out of 16 datasets and optimal ROC-AUC on 5.

Conclusions:

• The AVE hybrid model offers a more reliable and precise approach to outlier detection, especially for complex, high-dimensional data.
• AVE presents a robust and effective solution for real-world anomaly detection applications.