There has been much recent work on fraud and Anti Money Laundering (AML) detection using machine learning techniques. However, most algorithms are based on supervised techniques. Studies show that supervised techniques often have the limitation of not adapting well to new irregular fraud patterns when the dataset is highly imbalanced. Instead, unsupervised learning can have a better capability to find anomalous and irregular patterns in new transaction. Despite this, unsupervised techniques also have the disadvantage of not being able to give state-of-the-art detection results. We propose a suite of unsupervised and deep learning techniques to implement an anti-money laundering and fraud detection system to resolve this limitation. The system leverages three deep learning models: autoencoder (AE), variational autoencoder (VAE), and a generative adversarial network. We preprocess the given dataset to separate the Transaction Date attribute into its base components to capture time-related fraud patterns. Also, Wasserstein Generative Adversarial Network (WGAN) is used to generate fraud transactions, which are then mixed with the base dataset to form a more balanced mixed dataset. These two datasets are used to train the AE and VAE models. We built two versions of the AE model (single-loss and multi-loss) besides a novel method of calculating the anomaly score threshold, called Recall-First Threshold (RFT), which helps enhance the model's performance. Experimental results demonstrated that the False Positive Rate (FPR) drops down to as low as 7% in the proposed multi-loss AE model. In comparison, we achieved an accuracy of 93%, with 100% of the fraud transactions recalled successfully.
Narrow-based Exchange-Traded Funds (ETFs) offer significant diversification compared to stock market investments while maintaining a reasonable level of predictability through asset combinations and industry dependencies, rendering them ideal for the study of financial forecasting. In this research, a comparative analysis is presented focusing on Machine Learning (LightGBM), Deep Learning (LSTM), and Statistical (AR) methods for predicting the ETF market trends, with the autoregressive (AR) model outperforming across various evaluation metrics (RMSE, MAPE, R2). Beyond comparing different modelling techniques, the data aggregation approaches for constructing a multi-ticker model were discussed. The LSTM was greatly improved by the simple data concatenation, allowing complex training with a larger dataset, while the proposed shuffled-concatenation technique exhibits superior performance in LightGBM, surpassing all other models. These findings highlight the effectiveness of the multi-ticker model in improving prediction accuracy and emphasize the feasibility of forecasting shuffled time-series data using the tree-based model.
25 March 2025 in Intelligent Systems in Accounting, Finance and Management
In recent years, stock price prediction has increasingly become a pivotal research direction at the intersection of finance and computer science. While traditional forecasting models predominantly rely on historical prices and technical indicators, the advent of big data and artificial intelligence technologies has prompted researchers to integrate multi-dimensional data sources, such as news and social media, to enhance model performance. This study employs FinBERT for news sentiment analysis, combined with LSTM and GRU recurrent neural networks (RNN), aiming to elevate the accuracy of stock price predictions. Experimental results indicate that incorporating news sentiment scores significantly improves forecasting outcomes and particularly GRU demonstrates superior performance over LSTM. Notably, compared to large-cap stocks, investors in small-cap stocks exhibit heightened sensitivity to news sentiment.