Performance Comparison of Autoencoders for Anomaly Detection in IoT, Robotics, Unmanned Vehicles in Oil Industry Systems
Litsenziya: cc-by
Annotatsiya
Autoencoder-based models have gained popularity over the past few years due to their ability to learn effective representations of data using unsupervised learning. Traditional AEs, while being universally successful in many applications, tend to involve huge computational and memory requirements and are therefore not scal-able and cannot be deployed in real-life IoT systems. In a bid to overcome these challenges, the present study compares the Quantized Autoencoder (QAE-2024) model with a Sparse Autoencoder–Random Forest (SAE-RF) hybrid model for the purpose of improving computational efficiency and anomaly detection capabili-ties. The QAE-2024 model uses quantization methods to make the model compli-cated and compact and retain high reconstruction accuracy and powerful feature learning. The hybrid SAE-RF model brings in integration of sparse representation learning into robust classification power of Random Forests for improved accuracy and interpretability. Experimental evaluation was performed using the CIC-IoT 2023 benchmark dataset, which includes varied IoT network traffic patterns repre-sentative of smart and connected spaces, like those relevant to advanced material systems and robotic control. Results show that the SAE-RF model provides en-hanced classification precision, whereas the QAE-2024 model provides better computational efficiency and memory optimization. These results imply that au-toencoder hybrid and quantized structures provide viable solutions for real-time anomaly detection and optimization for IoT-based and Unamanned vechicles in oil industry and Robotic Systems, where precision, reliability, and effective data management are of utmost importance.
Hali tarjima qilinmagan