The financial services sector is on the eve of a major transformation, and the driving force behind it is AI. Innovative applications for AI have already been found across areas such as credit scoring, regulatory compliance, customer experience, and portfolio management. Thanks to rapid advancements in technology, tasks that once took employees hours to complete manually, can now be done in a matter of seconds.
One area that is growing exponentially is fraud detection, and with such large sums of money at stake, it is easy to see why. McAfee estimates that cybercrime, of which financial fraud is a component, now costs the world around $600 billion, equating to 0.8 percent of the global GDP.
Cybercriminals are getting smarter and, ironically, are leveraging advances in technology for their own benefit. Banks and financial institutions have no choice but to tighten their defences and develop their own capabilities faster.
According to Statista, in 2017, the global FDP (fraud detection and prevention) market was estimated to be worth $16.6 billion. Areas where fraud detection and prevention are applied include insurance claims, money laundering, electronic payments, and bank transactions, both online and offline.
AI and machine learning, combined with the rich data sets available in the financial services sector, are providing organizations with the means to protect their businesses, and defeat criminals. Let’s take a look at how.
Financial firms face a rapidly evolving threat from cybercriminals – attacks persistently come in the form of money laundering, identity theft, and mobile fraud, among others.
One of the most common types of cybercrime, however, is bank and credit card fraud. The growth in e-commerce and mobile payments is partly behind the soaring incidence of card fraud in recent years, and the scale of the problem is vast. According to McKinsey, worldwide losses from card fraud could be close to $44 billion by 2025.
As well as the direct cost of fraud, companies are also suffering because of lost sales when genuine transactions are declined by fraud management systems. McKinsey suggests that false positives constitute up to 25 percent of declined transactions for e-commerce retailers.
The challenge for banks and financial institutions then is to quickly identify and separate fraudulent transactions from those that are legitimate, without impacting on customer experience.
Traditionally, banks and financial institutions have approached fraud detection with manual procedures, or rule-based solutions, which have been limited in their success. A rule-based approach means that a complex set of criteria for flagging suspicious transactions has to be established and reviewed manually.
While this can be effective in discovering anomalies which conform to known patterns, it is not capable of detecting fraud which follows new, or unknown patterns. This gives criminals the incentive to develop ever more sophisticated techniques to circumnavigate the rules, and they themselves are leveraging new technologies to achieve this. The solution that is helping banks and financial institutions get one step ahead, is machine learning.
The high volume of transactional and customer data readily available in the financial services sector makes it ideal for the application of complex machine learning algorithms. Banks and financial institutions are able to automate the analysis of their customers’ behavioral patterns for any signs of abnormality, giving them the ability to identify and flag fraudulent activity in real-time.
As well as picking up known patterns, machine learning is able to go a step further and ‘learn’ new patterns, without the need for human intervention. This allows models to adapt over time to uncover previously unknown patterns, or identify new tactics that might be employed by fraudsters.
The increased accuracy of machine learning provides financial firms with a significant reduction in the number of false positives, where transactions are incorrectly flagged as fraudulent and declined, and false negatives, where genuine incidences of fraud are missed. Overall, firms are able to mitigate financial losses, as well as protect their reputations, maintain public confidence, and improve customer experience.
Given the complexity of financial fraud, and the speed at which cybercriminals adapt, a combination of supervised and unsupervised machine learning methods are needed to create a model with sufficient predictive capability and accuracy. Machine learning models operate with tens of thousands of parameters and are far more effective in finding subtle correlations in data, which may be hidden for an expert system, or a human reviewer.
In general, there are two types of machine learning algorithms used in fraud detection: supervised and unsupervised learning. The former uses already annotated data – reviewed and labeled as fraud activity by a human – to learn complex patterns in datasets provided by a business. The latter approach deals with datasets that have not been labeled and infers inner data structure by itself.
Data scientists have access to a range of techniques, which can be broken down in terms of problems they solve: classification and regression. Both can be used to analyse data and provide the answer to whether a transaction was genuine or fraudulent. The typical supervised machine learning algorithms used to solve these problems are logistic regression, decision trees, random forests, and neural networks.
Logistic regression is a popular method, which determines the strength of cause and effect relationships between variables in data sets. It can be used to create an algorithm which predicts whether a transaction is ‘good’ or not.
Decision trees can be used to create a set of rules that model customers’ normal behavior and can be trained, using examples of fraud, to detect anomalies.
Neural networks are a powerful technique inspired by the workings of the human brain. Able to learn and adapt to patterns of normal behavior, neural networks can identify fraud in real-time.
Unsupervised techniques are based on clustering algorithms, which group similar data points together – they are used for anomaly detection. Algorithms used in the unsupervised approach are K-means clustering, Local Outlier Factor and One-Class SVM.
K-means clustering divides a dataset into clusters. The algorithm works iteratively and assigns data points to one of the predefined number of classes (k), based on the features that are in the dataset. Data points are clustered based on feature similarity.
Local Outlier Factor, is an algorithm that calculates the local density of data points and allows for identifying regions with similar density in the data set. By using the locality concept, one can distinguish points with much lower density than other neighbours. These points are outliers (fraudulent transactions)
One-Class SVM learns a function used for novelty detection. The idea of novelty detection is to detect rare events, i.e. events that happen rarely, and hence, of which you have very little samples. The problem is then that the usual way of training a classifier will not work.
Although machine learning represents a huge leap forward compared to traditional methods of fraud detection, it is not without its limitations.
Machine learning models are only as good as the data they are provided with. While financial services have access to massive data sets, there are relatively few fraudulent transactions within these, which can reduce a system’s predictive capability. There are several approaches to dealing with this problem.
One way financial firms are overcoming the challenge of imbalanced data is artificially creating data points which imitate the under-represented class (oversampling). A popular technique is SMOTE (Synthetic Minority Over-sampling Technique). This technique uses the k-Nearest Neighbours algorithm to identify minority classes in the dataset and learns its features. Next, it randomly chooses one of the nearest neighbour data points and creates an artificial randomly changed data point near to it.
Another useful technique is undersampling. It’s a fairly simple algorithm, which randomly samples the dominant class to reduce its size.
The third approach, called the Combined Class Method, uses SMOTE to interpolate noisy data points on the boundary between outliers and inliers and the nearest neighbour technique to clean the data near the separatrix of both classes, making datasets cleaner and easier to distinguish.
By using these techniques, one ensures that the data fed to a machine learning algorithm are much better balanced and more resistant to overfitting.
In addition, the current uses of machine learning in fraud detection still involve a two-stage process where the first stage is automated, but the second stage requires manual checking. Such a system exposes financial firms to the possibility of human error or deliberate misuse. Organizations must move towards complete automation to fully realize the benefits of employing AI.
As technology advances and criminal schemes become more sophisticated, banks and financial institutions are leveraging the power of AI to protect their businesses and improve customer experience.
Applying machine learning to fraud detection enables financial firms to identify genuine transactions versus fraudulent transactions in real time, and with greater accuracy. Through a combination of supervised and unsupervised methods, models are capable of learning and recognizing new patterns that may have been missed by other approaches to fraud management.
Given the huge sums of money at stake, and the relentless threat posed by criminals, employing cutting-edge systems with the ability to learn will become essential as companies strive to stay ahead. While organizations may not be in a position to move to advanced data analytics immediately, they should begin examining their existing data, identifying data requirements, and developing the expertise necessary to begin as soon as possible.