A Hybrid LSTM-KNN Framework for Detecting Market Microstructure Anomalies:

This paper proposes a novel hybrid LSTM-KNN framework for detecting market microstructure anomalies in high-frequency credit default swap (CDS) markets. The framework integrates the temporal learning capabilities of Long Short-Term Memory networks with the pattern recognition strengths of K-Nearest Neighbors classification to identify price jumps and market anomalies. Through analysis of high-frequency CDS market data spanning from 2020 to 2023, encompassing over 2.5 million data points from five major CDS indices, the research demonstrates significant improvements in jump detection accuracy. The hybrid model achieves a 92.8% accuracy rate, representing a 15.2% improvement over traditional statistical methods and an 8.5% enhancement compared to standalone deep learning approaches. The framework maintains computational efficiency with an average processing latency of 48.2 milliseconds, enabling real-time market applications. The empirical analysis reveals strong correlations between detected jumps and market liquidity conditions, with bid-ask spreads and order book imbalances identified as critical predictive indicators. The research contributes to both theoretical understanding of market microstructure dynamics and practical applications in risk management and market surveillance.