This paper introduces a novel nonparametric hybrid cyber-intelligence-based statistical process control and anomaly detection framework in time series data. It is developed to overcome the shortcomings of the classical control schemes in dealing with complex, abnormal, and noisy input data, especially when it is autocorrelated. The proposed methodology combines three technical pillars: First, it utilizes a bidirectional long-short-term memory architecture (Bi-LSTM) to capture long-term time dependency and learn nonlinear patterns, leaving only true deviations as residuals that remove trends and noises from the market. Second, it adopts the Golden Eagle Optimizer (GEO) algorithm for optimal parameter selection. This intelligent algorithm tunes the smoother factor (l) and the control boundary (L) at a certain sample size to minimize the Average Run Length (ARL) of the nonparametric exponentially weighted moving average (NPEWMA-SR) scheme. Third, the framework is validated via R software. The framework was applied to Google's daily trading data using different sample sizes (10, 30, 60, 120, 250, 365, 600, 900, and 1245) days of 2026, to detect the shift in the system, within 2 trading days, achieving an In-control Average Run Length ARL0 = 499.6 and an Out-of-control Average Run Length ARL1 = 1.65 days. The system demonstrated high statistical stability, a very low false alarm rate, and the best statistical sensitivity among all sample sizes. These results prove its effectiveness across small, medium, and large samples, making it a powerful early warning system for monitoring market volatility.
