Machine Learning in RegulatoryCompliance Software Systems : An Industrial Case Study

Abstract: The presented study investigates the role of Machine Learning in Regulatory Compliance software systems by conducting a case study in the maritime industry. There is a need to obtain a clear understanding of Machine Learning approaches for automating regulatory compliance.  Background: Organizations and software developers face different regulations in different domains that need to be satisfied by the requirements of systems. Trading across borders brings challenges in managing each country's regulatory requirements. Even though a large amount of effort from international and local organizations, an efficient AI-based system is not yet fully deployed for offering decision support in handling such complex issues as a viable solution. The increasing enforcement of sanctions and anti-money laundering by various institutions and countries poses a significant threat to many industries worldwide if they are not educated or aware of the legal and financial risks. therefore, Regulatory Compliance plays a critical role in this area. Regulatory Compliance is a guideline for laws or regulations related to the business that the stakeholders must abide by to promote a safer business environment and benefit society. Developing a Digital Compliance(DigiComp) system as a software service is a solution that involves both technical and organizational challenges which need a large amount of research. As a case study, an AI-based DigiComp system has been implemented in the port of Vordingborg for the EU-financed project "Connect2SmallPorts". A major Corporate challenge was the choice of appropriate and efficient approaches and tools for designing the automated regulatory compliance system. This study investigates the role of Machine Learning in regulatory compliance systems and its possible opportunities and limitations.  Objectives: The significant role of compliance management made it an interesting research topic since some highly regulated systems need to follow specific laws and guidelines. As the objectives of this study, the role of machine learning on regulatory compliance systems has been investigated by extracting the most popular ML approaches and their benefits and challenges from previous studies and then analyzing the related benefits and limitations in a real-world case. Methods: We have done a systematic literature review as the qualitative research method with database search and snowballing to collect the applied Machine Learning approaches and their benefits and challenges for Regulatory Compliance software systems. Then a case study was conducted for implementing a regulatory compliance system in the maritime industry and investigating the benefits and challenges in a real-world project. Also, Focus groups and interviews with the stakeholder and domain experts were held as data collection methods during the case study.  Results: After investigating the existing challenges in regulatory compliance and identifying the risk assessment framework based on Machine Learning as the most popular AI approach in this area, we implemented a risk assessment framework based on neural networks. It provides high accuracy and a low error rate in predicting the future state to prevent the non-compliance risk. Conclusions: Examining the proposed digital compliance system explores its similarity to the benefits and limitations extracted from SLR. This study has provided the development of a new AI-Based system that can inspire software companies to build more efficient regulatory compliance systems for different domains in the industry.