Title: Stock market manipulation detection using feature modelling with hybrid recurrent neural networks

Authors: Sashank Sridhar; Siddartha Mootha

Addresses: Department of Computer Science and Engineering, College of Engineering Guindy, Anna University, Chennai, TN, India ' Department of Computer Science and Engineering, College of Engineering Guindy, Anna University, Chennai, TN, India

Abstract: A stock market is a potent platform which handles a large of number of transactions within a second. Keeping track of every single transaction is a daunting task for regulatory bodies. The objective of a regulatory body is to ensure a fair trading environment and to verify that the price of a stock is not being manipulated. This paper proposes a hybrid stacked artificial neural network and recurrent neural network to model the static and dynamic features of stock data. Based on the manipulated stocks, affidavits provided by the Securities and Exchange Board (SEBI) of India, a daily trading dataset is created by scraping the Bombay Stock Exchange (BSE) website. The system is capable of identifying three types of manipulation scenarios. The proposed hybrid system is compared to various supervised algorithms, and various ensemble models and the system outperforms all with an accuracy of 96.06%.

Keywords: manipulation detection; hybrid neural networks; ensemble learning; recurrent neural networks; RNNs; fraud detection; long short-term memory; LSTM; bidirectional long short-term memory; Bi-LSTM; stacked generalisation; artificial neural networks; ANNs; feature engineering.

DOI: 10.1504/IJNVO.2022.121864

International Journal of Networking and Virtual Organisations, 2022 Vol.26 No.1/2, pp.47 - 79

Received: 31 Aug 2020
Accepted: 26 Apr 2021
Published online: 07 Apr 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article