Multilayer one-class extreme learning machine.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• One-class classification is crucial for anomaly detection.
• Existing methods like one-class SVM and autoencoders struggle with complex, multi-class data.
• One-class extreme learning machine (OC-ELM) offers faster learning but limited effectiveness.

Purpose of the Study:

• To develop advanced one-class classification algorithms for complex data.
• To improve feature representation and generalization performance in anomaly detection.
• To introduce the Multilayer Neural Network based One-Class Classification with ELM (ML-OCELM) and its kernel-based variant (MK-OCELM).

Main Methods:

• Utilizing stacked autoencoders (AEs) within ML-OCELM for effective feature representation.
• Investigating a kernel-based learning framework in stacked AEs for MK-OCELM.
• Employing multilayer neural networks and extreme learning machine principles.

Main Results:

• ML-OCELM and MK-OCELM demonstrate superior performance compared to OC-ELM and other state-of-the-art algorithms.
• Experiments on 13 UCI benchmark datasets and urban acoustic classification validate the proposed methods.
• MK-OCELM shows reduced parameter dependency and strong generalization capabilities.

Conclusions:

• The proposed ML-OCELM and MK-OCELM effectively address the limitations of existing one-class classification algorithms.
• These novel methods offer enhanced anomaly detection for complex and multi-class classification tasks.
• The research highlights the potential of deep learning architectures combined with ELM for robust outlier detection.