The Simility Blog
Adopt Clear-box Machine Learning to Make Explainable Decisions
Jayan TharayilMarch 07, 2019
Clear–box machine learning algorithms allow businesses to make informed decisions, while accurately detecting fraud.
Machine learning is playing a pivotal role in enhancing customer experience and helping businesses fuel growth. It has evolved as the go-to-technology when it comes to fighting fraud. This is largely because machine learning algorithms have an incredible ability to unearth patterns from seemingly unrelated data and help businesses better detect fraud. Machine learning becomes even more incisive with larger data sets, allowing businesses to make informed decisions in real time.
Black-box, Opaque Machine Learning
With the backdrop of machine learning holding immense potential, it may come as a surprise that this technology may not be all that fair as it is considered to be. This is because we all know machine learning algorithms are math-based and involve statistical equations. But, according to the principle of ‘GIGO’ (garbage in garbage out), machines are only as good as the data they are fed. Machine learning systems learn behavioral patterns from training data to make decisions and create algorithms for future decision-making.
Customers are increasingly becoming dependent on numerous data-driven services and payment options. While this bodes well for digital businesses in today’s data-driven economy, it also opens up concerns about intentional or implicit biases emanating from both the data and algorithms, and how they are used in decision-making. Therefore, building transparency and fairness into machine learning poses a big challenge.
Building in Transparency and Explainability
Using additional data and monitoring the effects of decisions made over a period of time can be useful in mitigating biases and building in algorithmic transparency. That said, most of the current machine learning systems are not easily explainable due to a black–box approach. Clear–box machine learning or transparency in algorithms, refers to the ability to see all the steps in the process—from collecting data sets to the final result, which makes interpretation easy. When businesses can see for themselves the steps that a platform or an algorithm takes in order to transform data into results, they are more inclined to trust and use the solution.
When machine learning systems are transparent, decisions are backed by reasons. Clear–box machine learning models not only enable decision-making, but also provide the ‘why’ behind decisions. As a result, businesses can be confident of the decisions they make. Due to increased transparency, businesses can efficiently govern machine learning algorithms and ensure that the decisions are aligned to the business’ objectives. Businesses can also modify the rules, and enforce greater checks on the algorithms and training data to ensure sustained value. Besides transparency, clear–box machine learning also helps bring in accountability, since the systems can be completely audited.
Better Fraud Prevention
Clear–box machine learning with supervised models is more efficient in fighting fraud when compared with black–box machine learning because it can unlock unprecedented insights to investigate red-flagged transactions. Using intelligence from varied data sources, these algorithms can accurately detect fraud while allowing genuine customers to seamlessly transact. And with the growing sophistication and scale at which cybercriminals are orchestrating attacks, clear–box machine learning tools with adjustable rules are the way forward.
Simility’s Clear-box Approach
Simility, a PayPal service, uses both supervised and unsupervised machine learning algorithms in its Adaptive Decisioning Platform that are designed to help businesses accurately detect and block fraud. This omni-channel platform can ingest data—both structured and unstructured—from varied sources to provide actionable insights in real time. It enables businesses to implement strong governance around machine learning strategies, including hypothesis testing and champion-challenger based machine learning processes to explain why the system made a decision. Thus, with the ability to monitor results, and modify rules and models, businesses can create smarter, more incisive fraud strategies that can adapt to the evolving threat landscape. Moreover, strong visualization on top of a fraud-centric data lake helps identify, conceptualize, validate, and operationalize fraud intuitions more quickly.
To learn more about the clear-box machine learning features of the Adaptive Decisioning Platform, schedule a demo now.