Abstract. This study explores a comprehensive approach to analyzing laser power stability by combining statistical evaluation with machine learning-based predictive modeling and anomaly detection. Power data from an Erbium-doped femtosecond fiber laser operating at 775 nm are analyzed to assess variability, trends, and potential instabilities. Statistical analysis revealed moderate fluctuations in power output. Advanced anomaly detection techniques, including Isolation Forest and K-means clustering, identified distinct deviations in the data, with K-means achieving a Silhouette Score of 0.73. Predictive modeling using linear regression and ARIMA demonstrated robust forecasting capabilities. The ARIMA model effectively captured both short-term fluctuations and long-term trends, projecting stabilization of laser power over a 300-minute extension, indicative of equilibrium behavior. This study highlights the integration of statistical and machine learning tools as a valuable framework for enhancing precision and stability in high-performance laser applications.

Key words: Laser power, Laser power stability, anomaly, K-means clustering, Isolation Forest, linear regression, ARIMA model, predictive modeling, statistical analysis, power stability index (PSI), coefficient of variation (CV), machine learning in laser systems, time series forecasting.