Publication Details
Abstract
This study examined the effectiveness of Artificial Intelligence (AI) in detecting financial statement fraud in Port Harcourt Metropolis, Nigeria, with a focus on Natural Language Processing (NLP) techniques and Model Interpretability frameworks. Using a descriptive research design, data were collected from 400 professionals across 25 commercial banks, two microfinance banks, and one specialized financial service provider. Key stakeholders included internal auditors, risk management teams, financial analysts, senior bank executives, data scientists, and regulatory bodies. A purposive sampling technique was employed, and data were gathered through structured questionnaires and semi-structured interviews. Findings revealed that AI-driven fraud detection enhances audit credibility, risk identification, and financial transparency. Specifically, interpretable AI models, such as SHAP and LIME, significantly improved auditor confidence and fraud risk assessment. However, concerns were raised about the reliability of some NLP techniques, such as text mining. Descriptive statistics (Mean and Standard Deviation) were used to analyze survey responses, while content analysis was applied to interview data. Reliability was assessed using Cronbach’s Alpha (≥ 0.7), and test-retest reliability confirmed instrument consistency. The study concluded that balancing AI model accuracy with interpretability is crucial for fraud detection effectiveness. It recommended that banks and regulatory bodies invest in explainable AI models and provide continuous training for financial professionals to enhance their understanding and application of AI-based fraud detection. The findings highlight the growing relevance of AI in financial fraud prevention, emphasizing the need for transparent and trustworthy AI solutions in the banking sector.