Ignite Machine Learning with Decision Rules | Risk Management

Chalkboard with Machine Learning on it

When today's businesses create a list of must-haves in the race to alter and ultimately transform how they operate, machine is one of those checkbox necessities. And indeed, when you run the gamut of business-consumer touchpoints, machine learning is, indeed, one of the mainstays.

What's machine learning, and why is it needed? Machine learning is powered by algorithms that learn from data, identifying critical patterns, connections and insights – without being programmed to reach specific conclusions.

Machine learning enables more data of multiple types to be analyzed quickly, in order to derive maximum benefit from it. It takes advantage of exponential boosts in data storage capabilities and computing power (at a fraction of the cost than what was previously possible) to help make , smarter decisions while constantly learning, evolving and providing unexpected new insights.

Where machine learning models are superior to traditional approaches

One of the core benefits of machine learning is boosting the predictive power of the models that businesses use to make better decisions. Here is an example.

Consider the risk modeling scorecard, which has been traditionally lauded by businesses and regulators alike for its predictive capabilities. In the scorecard environment, customer segmentation is based on “hard” lines and broad categories, such as new customer vs. existing customer, or high, medium and low risk. This paradigm, however doesn't capture the behavior of certain individual entities or provide more optimal ways to segment scoring models.

This is where machine learning can help you do a better job of predicting behaviors, and make faster decisions.

FICO machine learning capabilities have been around for more than 25 years, initially in fraud and credit risk, and extending to other operational and customer lifecycle use cases. FICO analytic tools and core applications use machine learning-powered techniques such as collaborative profiles (defined as distilling behaviors from a large group down to a few basic behavioral “archetypes”) to reveal entity segmentation based on customer behaviors. This approach groups customers into micro-segments based on that similarity versus typical segmentation approaches that rely on hard business attributes. Consider, for example, collaborative profiles that derive behavioral archetype distributions — these could include archetypes that point to credit seekers building credit histories vs. those who have higher risk and covering misuse of credit elsewhere in their history.

You can teach models to discover maximum predictive power, and find new relationships amongst input features that could produce a stronger model. For example, utilization is a key feature in a credit model, as is delinquency, but a nonlinear combination of these can produce better (sometimes, substantially better) results in a machine learning model. You can then drive these new inputs into a traditional scorecard model to ensure explainability.

How do decision rules boost machine learning?

Perhaps data science and analytics (and optimization – see how one loyalty program is pulling it all together) seem like the primary mechanisms to turbo-boost machine learning. But you still need ignition to harness all that power. That's where decision rules are your keys, figuratively and literally. Consider some key areas where machine learning benefits from rules.

First, decision modeling unlocks the true value of data in an unexpected way. Decision rules can help you prepare data for machine learning. Actually, decision modeling does this by putting the decision before the data, versus the other way around (and if you really want to dig deeper, check out this white paper). A decision-centric approach defines the business objective first and builds the decision model to achieve it. The analytic and data requirements are defined by the decision model, so data exploration and analytic development are faster and more productive. Tools like Blaze Advisor and Decision Modeler are built for this approach; however, even the best technologies alone aren't enough. FICO's Dr. Alan Fish developed decision requirements analysis (DRA) to integrate – leveraging visual tools based on the Decision Modeling and Notation (DMN) standard. The core driver of DRA is a workshop that helps businesses construct a model of the domain of decision-making: Namely, covering the activities required to discover, document, develop and maintain decision rules that will help businesses better position for next-generation-now capabilities such as machine learning.

Automate and operationalize your machine learning-powered algorithms. Blaze Advisor and Decision Modeler are optimized to help organizations to automate high-volume operational decisions quickly and effectively while maintaining business control over those decisions. This is particularly critical in today's business environments that require both agility and consistency to boost performance and customer-centricity. Many businesses are taking automation to the next level with machine learning. Machine learning helps you make smarter, more precise and data-driven decisions while rules provide the infrastructure to operationalize your algorithms and models (i.e., activate machine learning for your decision services and applications). This combination can literally reshape how businesses operate. Also note that decision requirements analysis, mentioned above, is essential to help define what decisions to automate, vs. those that require human intervention (or any combination of the two).

Data science and decision rules

Great, where do we start?

Machine learning is but one element of business transformation that benefits from a decision-first approach. Businesses that FICO works with have adopted this approach to often initially address specific business challenges, and then evolve over time to larger decision centralization or transformation initiatives. Within the DRA process discussed earlier, subject matter experts in areas such as risk, marketing or other areas will define the business decisions to be made, but other stakeholders – including IT, data scientists, operations researchers, and even senior-level executives – may weigh in on specific elements of the effort. Decisions include (but are not limited to) elements such as rule sets, scorecards, decision tables as well as – you guessed it – machine learning algorithms (typically, within analytic and/or optimization models).

Whether you're running a line of business, part of IT, engaged in data science or other analytic areas, or creating and managing decision rules, FICO has a rich and varied set of content around machine learning-related capabilities:

CxO and LOB Leader perspectives

  • White Paper: Why a Decision-First Approach Is Critical for Competitive Advantage
  • Case Study: Loyalty Is Rocket Science for a Major Canadian Grocer
  • Hot Topic: Accelerating Decision Automation Time to Value with Decision Requirements Analysis
  • : Can Machine Learning Save Big Data?

Analytic and Decision Rule Professional perspectives

  • White Paper: 12 Secrets of Business Rules Success
  • Blog: Blaze Advisor 7.5 – The Final Frontier
  • Blog: Scorecards and Machine Learning – Replace or Embrace?
  • FICO Decision Modeler Trial

To stay on top of the latest Blaze Advisor and Decision Modeler innovations, including (but not limited to) machine learning, join our community.

The post Machine Learning with Decision Rules appeared first on FICO.

You might also like
Leave A Reply

Your email address will not be published.