My recent job has not only transformed my perspective on work but also challenged my prowess in data analysis. I’m employed in the Financial Fraud Risk Analysis department of the Personal Wealth Management division under Hang Seng Bank. The core responsibility of this role is to assist in identifying and mitigating risks associated with financial fraud. Financial fraud encompasses a diverse array of scenarios, ranging from corporate money laundering and fraud to individual offshore funds and forex scams.
My colleagues, superiors, and I are tasked with monitoring and assessing every transaction involving personal wealth investments, aiming to identify transactions that carry potential risks. This demanding role requires a vigilant eye and an adept understanding of financial patterns. The ultimate goal is to safeguard our clients’ financial interests by promptly addressing and preventing fraudulent activities.
This position has not only altered my perception of work dynamics but has also tested and further honed my data analysis skills. The intricate task of detecting subtle irregularities within financial data has reinforced the significance of meticulous analysis and its pivotal role in maintaining the integrity of financial systems.
In most instances, the job itself isn’t overly complex, as each investment and transaction follows strict protocols and undergoes initial security and risk assessments by frontline banking professionals. Therefore, when it’s my turn to review transactions, the majority of them are legal and compliant. However, during my tenure, a significant event occurred within the company—an event that could be described as a high-risk endeavor—the merger of clients.
In this context, I found myself unexpectedly in the spotlight, owing to the urgent need within our department to swiftly review vast amounts of customer information, coupled with the inability to access the company’s internal risk management system. My prowess in data analysis gained significant recognition. I emerged as the standout data analyst in our team, naturally taking on a more prominent role. This situation translated into increased responsibilities and access to additional resources. However, this elevation came with a trade-off: the task of accurately identifying high-risk clients within the merger pool within the designated time frame.
Due to the bank’s information protection agreement, my code will be shown in another bolg on the homepage with similar code of my own creation
The initial goal of this project was straightforward: to utilize existing labeled data within the company to construct a predictive model. This model would then be applied to new customer data resulting from mergers, allowing us to identify high-risk clients. In this context, my responsibilities boiled down to two key tasks. Firstly, I needed to cleanse the data of these new customers, transforming it into a format compatible with our company’s risk data. Secondly, I had to select and build a model. Given that this predictive model didn’t require real-time predictions but demanded heightened accuracy, I opted to prioritize precision.
The data cleansing aspect actually consumed a substantial portion of my time and energy. The datasets from the two companies presented vastly different data types and wide variations in features for each customer. This necessitated extensive coordination, particularly in aligning data between the two entities. The challenge lay in reclassifying features based on our standardized categorization methodology.
To overcome this hurdle, I engaged in direct communication with counterparts responsible for customer information in the other company. Our goal was to establish a mutual understanding and have them reclassify their features in line with our categorization. This collaborative effort aimed to harmonize the data and enable seamless integration into our existing risk assessment framework.
This process involved in-depth discussions and consultations, not only with my immediate team but also with colleagues from our company’s customer information department. Through these interactions, I strived to ensure that the newly acquired data adhered to our enterprise’s feature classification standards. This concerted approach was crucial in facilitating accurate data analysis and establishing a foundation for robust predictive modeling.
Subsequently, after careful model selection, I arrived at the decision to employ the XGBoost model for predictions. Additionally, I explored the use of the Random Forest model to interpret the results of the original data. This dual approach enabled me to not only predict high-risk clients effectively but also to uncover the contributing factors underlying these predictions.
This methodology led to a strategic approach in presenting my findings to our superiors. I first elucidated the reasons behind the predictions of high-risk clients using the Random Forest model for the existing data, providing the management with insights into how the model interprets risk factors. Armed with this understanding, I then outlined my predictions for the new clients resulting from the merger, contextualizing these predictions within the framework of the company’s existing risk landscape.
This structured approach facilitated a comprehensive understanding of the potential risks and instilled confidence in my findings. By connecting the predictive models with the business realities, I ensured that the insights generated were not just technically sound but also directly applicable to the company’s decision-making processes.
