AI Summary & Key Takeaways
AI InsightThese results once again demonstrate how effective hybrid models can be in microservice performance monitoring processes.
📌 3 Key Takeaways
Abstract
ABSTRACT Microservice architectures have become one of the fundamental building blocks of distributed systems today. However, the complex structure of these systems creates significant challenges in areas such as performance monitoring and anomaly detection. In this study, we developed a hybrid model based on autoencoders and Convolutional Neural Networks (CNN) to detect and classify anomalies on a simulated microservice dataset. The autoencoder identifies anomalies by learning normal performance behaviors, while the CNN classifies these anomalies into “Normal,” “High Load,” and “Weak” categories. In our experiments using a simulated dataset containing 10,000 examples, our model achieved a high classification accuracy of 98.8%. The autoencoder's anomaly detection performance was quite satisfactory, with 0.728 precision, 0.990 sensitivity, and 0.840 F1-score. Our hybrid model delivered successful results with F1-scores of 0.988 and 0.993 in the “Normal” and ‘Weak’ classes, respectively, while showing a slight limitation in the “High Load” class with an F1-score of 0.945 due to class imbalance. These results once again demonstrate how effective hybrid models can be in microservice performance monitoring processes.