Akademik Veri Yönetim Sistemi
El-Cezeri Journal of Science and Engineering, cilt.13, sa.1, ss.101-111, 2026 (Scopus)
The modern microservice architectures require agile configuration management to ensure optimal performance and reliability. Nevertheless, the dynamic and distributed nature of these systems makes it difficult to assess how configuration changes affect key performance indicators (KPIs) such as CPU utilisation, memory consumption, response time and error rates. In this article, we propose a time-series driven approach to analyse and optimise the impact of configuration changes in large-scale enterprise environments. As a first step, using correlation and regression analyses, we quantify the causal relationships between critical configuration parameters (cache size, thread pool size, and release complexity) and performance metrics. Then we apply time series modelling techniques (Prophet, ARIMA and an LSTM-based Autoencoder) to detect anomalies caused by misconfigurations and traffic fluctuations. The LSTM based Autoencoder outperforms traditional anomaly detection methods by achieving 22% higher sensitivity in detecting performance deviations. The system performance is further improved by integrating Bayesian Optimisation and reinforcement learning methods for automatic parameter tuning, showing up to 25% reduction in response times and 17.8% improvement in CPU utilisation. In addition, this work investigates deployment strategies such as blue-green and canary releases and their interaction with access processes. These findings underline the importance of data-driven configuration management for microservices and provide actionable insights to achieve increased system stability, lower operational costs, and rapid recovery from performance anomalies.