An Integrated Model for Stock Manipulation Detection: Incorporates Low- and High-Frequency Data and Combines Supervised and Unsupervised Learning

Stock manipulation remains a significant challenge in maintaining market integrity and protecting investor interests. This paper introduces a robust integrated model for detecting stock manipulation, combining both supervised and unsupervised learning techniques and leveraging high-frequency and low-frequency data. The model employs a novel approach by integrating Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) networks, supplemented by Convolutional Block Attention Modules (CBAM) and an Adversarial Variational Autoencoder (A-VAE) to enhance the detection capabilities. We extensively test the integrated model based on the historical manipulation cases published by the China Securities Regulatory Commission. Experimental results demonstrate superior performance, accuracy, and recall of 84% and precision of 86%, indicating high performance and reliability in detecting stock manipulation activities. The findings highlight the effectiveness of combining multiple data types and learning strategies, offering a valuable tool for regulatory bodies and market participants.