Chatter is considered as one of the important factors affecting the quality of machining processing, yet traditional chatter monitoring algorithms often lack sensitivity to chatter onset and struggle with real-time adaptability in setting monitoring thresholds. To tackle this challenge, we propose a self-adaptive online monitoring method for early chatter identification. The use of the improved wavelet packet energy entropy ( IWPEE) algorithm enhances chatter feature extraction, thereby improving recognition accuracy, robustness, and efficiency. Moreover, an improved Pauta criterion dynamically determines the chatter monitoring threshold, enabling adaptive threshold calculation under varying processing conditions. Subsequently, we develop online chatter monitoring software to meet the practical monitoring demands of machining. Validation of the proposed algorithm through simulation signals and cutting experiments demonstrates a 360% increase in sensitivity compared to traditional entropy-based methods. Additionally, the adaptive determination of the threshold by the improved Pauta criterion ensures successful monitoring of chatter onset during its growth stage. Furthermore, significant enhancements in threshold stability and adaptability relative to traditional threshold algorithms are demonstrated.