Machine learning is currently the most well-known form of artificial intelligence. It provides systems with the ability to automatically learn and improve from experience, without the need for explicit programming and with minimal human intervention.
Machine learning uses algorithms to receive and analyse input data to identify patterns and predict output values. Most importantly, as new data feeds into the algorithms, the machine learning model can learn and optimise its performance, gaining intelligence over time.
This article will give you a brief introduction to machine learning and how it is used for AML compliance.
From healthcare to financial services, machine learning is used across a wide range of sectors and for many different purposes. Some of these include:
For anti-money laundering (AML) and counter terrorist financing measures (CFT) in particular, Financial Action Taskforce (FATF) recommends the use of machine learning to help implement a risk-based approach and increase the degree of confidence had in such measures.
Machine learning can allow for real-time, quick, and more accurate data analysis. FATF advocates for the use of machine learning to automate the process of risk analysis, allowing the machine to take account of a greater volume of data and to identify emerging risks which do not correspond partially or fully to already-understood profiles.
According to FATF, machine learning can add most value in the following areas:
Machine learning uses anomaly detection algorithms that are commonly categorised according to whether they are supervised or unsupervised. These types of machine learning algorithms can be described as:
Supervised machine learning algorithms learn by example. They are provided with a known dataset that includes the desired inputs and outputs. Using this data, they must find a way to arrive at those inputs and outputs and make predictions. The operator will correct the algorithm as necessary until it reaches a high level of accuracy.
With unsupervised machine learning, algorithms study data to identify patterns that are not necessarily known. Unsupervised machine learning mainly consists of spotting unusual data correlations or unexpected customer behaviours. The operator does not provide instruction or guidance.
Semi-supervised machine learning is like supervised learning but uses both labelled and unlabelled data. With the support of the labelled data, the algorithm can learn to label the unlabelled data.
These algorithms learn from trial and error, and ultimately operator feedback. This regimented learning process provides the machine learning algorithm with a set of actions, parameters, and end values to which it must adapt its approach to achieve the best possible outcome.
Machine learning is an application of artificial intelligence (AI), a term which refers to a computer automating tasks that would otherwise be performed by a human.
For machine learning to work effectively, it is essential that the data supply is good in terms of both quality and quantity. Machine learning is entirely dependent on structured, semi-structured, and unstructured data. With no data, there is no machine learning.
To get started, every new machine learning system will first need to learn from an organisation’s historical data. The task of preparing data for machine learning should therefore not be underestimated.
As well as internal sources (such as account activity and transactions), data will need to be incorporated from external sources (such as sanctions lists and credit ratings) to provide insights and improve the system’s accuracy.
Management of data should be a company-wide activity whereby data capturing, cleaning, integrating, curating and storing is integral to all operations. Reflecting just how important data is, the preparation and cleaning of data takes so long that data scientists only have about 20% of their time left over to spend on analysis and creating insights.
All this means that before a system that uses machine learning can be fully deployed, it will first need human input to fine-tune its initial ‘self-taught’ learning (see image below).
Human input allows the machine to learn what anomalous behaviour looks like, and in turn correctly identify it when it occurs. Machines need to learn from both low-risk/normal behaviours and high-risk/abnormal behaviours.
Did the system correctly identify criminal activity (a true hit)? Or did the transaction pattern transpire to be normal behaviour (a false positive)?
The biggest challenge here is that machine learning is not a correct and forget exercise. In the financial sector, criminals constantly change their tactics to remain under the radar. For this reason, much of a machine’s learning and improvement will occur once the system is live, as a result of humans reviewing the alerts generated by the system and providing feedback.
When analysts feed their knowledge back into the machine, the machine can improve the accuracy and relevance of its alerts to reduce false positives and avoid false negatives in the AML programme.
AI and machine learning are often confused because machine learning is a term that is commonly used alongside or instead of artificial intelligence. Machine learning is an application or sub field of artificial intelligence. While AI and machine learning are often confused, AI comprises additional disciplines and areas such as advanced heuristics, decision management and computer vision.
Napier incorporates machine learning applications into its products by using supervised and unsupervised learning to:
Napier’s machine learning complements and enhances a rule-based approach to screening, supplying additional insights into unusual correlations. By drawing on historical data, the machine highlights the most relevant anomalous data points for analysts to review.
Napier’s AI Advisor is an optional feature within Napier’s screening solutions that identifies false positives to help analysts review alerts more efficiently. By using machine learning to analyse screening outcomes and improve match scoring, it can determine if the match should be discounted or if it requires further review.
AI Advisor works by scoring each match, showing the components that contributed to the score in a clear visual on the screening dashboard to help analysts make quick decisions about the quality of the match. AI Advisor provides an explanation alongside the score to help analysts understand the key factors in its decision: why a match was created and what was unusual about it.
The additional insights from AI Advisor, which analyses multiple additional variables to score a match, can help reduce false positives by up to a further 40%.
Discover Napier’s Intelligent Compliance AML solutions. These advanced solutions use machine learning for Transaction Monitoring and Transaction Screening, allowing you to easily enhance your analysis.
Learn more about machine learning here.