Sign In

Pattern-Driven Analytics: The Mirage That Misleads

2 minutes

Pattern recognition is a powerful human capability, but it can be misleading. We routinely detect patterns that are not real, and when these are embedded in analytics, the consequences can be severe. Given enough data, patterns seem to appear everywhere, from sales projections to flu seasonality, but whether these patterns are real or illusory is often unclear.

Humans naturally seek patterns in everything they see from ancient Greeks seeing constellations to modern analysts seeing patterns in data.

Pattern Validation

A purely data-driven approach to validating patterns lies at the heart of many analytics failures. Practices such as data mining and investigative data deduction often do little more than dress randomness in quantitative language.

Deductive pattern validation starts with observing a pattern then trying to prove its validity. This can be easily done using enough data and statistical manipulation and is really just post hoc story telling.

Patterns add value only when validation starts with a clear hypothesis, is tested using predefined methods, and is confirmed through external information and expert judgment. Analytics are important and this level of rigor is necessary for reliable results.

Inductive pattern validation starts with a hypothesis that is tested against a pattern observation and followed by comparing to external information and consulting a problem expert. This is the scientific method and is robust.

Problem Decomposition

Proper pattern verification works best when complex problems are broken into simpler ones. Advanata is designed to make this decomposition fast and easy.

Problem decomposition takes a high complexity problem and decomposes it into lower complexity problems. This makes it easier to hypothesize underlying mechanisms, makes more external information available, and improves engagement with the problem expert. Advanata is built on a directed acyclic graph structure that supports this decomposition process.

Decomposing a Complex Problem

The following example shows how breaking a problem into simpler parts allows for improved pattern verification.

A conglomerate would like to maximize their overall profitability and must measure the expected profit rate of one of its hotels. This is a high complexity problem that can be decomposed into smaller lower complexity problems such as room occupancy rate, fixed costs, seasonal events...etc.

Applying a deductive approach to the problem offers no reliable way to separate real patterns from random noise.

The deductive approach tries to immediately solve the high complexity problem of expected hotel profit rate. If a pattern is observed, it is self-justified and is accepted as the result with no further external challenge.

Using an inductive approach on one of the simpler problems enables reliable pattern verification and produces a better result.

The inductive approach would take one of the less complex problems such as room occupancy rate and upon observing a pattern it would be tested against the hypothesis, compared to external information, and confirmed with the problem expert. The consensus result is rigorously tested and is well validated.

Don’t Trust the Pattern

Distinguishing real patterns from random noise is essential to performing true analytics rather than pseudo-quantitative storytelling. Proper safeguards must be in place to ensure that verification is thorough and reliable.

If we observe a pattern we have to make sure we have external references and then seek consensus against our original hypothesis, external information, and the problem expert. If no references exist, then we should reject the observation.

The pattern in the first image comes from a random number generator, showing how easily false patterns emerge. A proper methodology for pattern verification is critical, and Advanata‘s underlying structural design simplifies problem decomposition to support accurate verification.

Sign up for early access to articles, delivered every two weeks. Please check your spam folder for the confirmation email.

Ready to see Advanata in action? Explore more resources, read related articles, and access the full application at advanata.com.

Want a personalized walkthrough? Contact info@advanata.com to get application access and schedule a demo.

Leave a comment