Rule-based systems vs. Machine Learning

Rule-based systems vs. Machine Learning

We can safely predict that in the next 5 years most of the rule-based insurance systems will be replaced by more dynamic Machine Learning based systems for claims handling and WAF detection.  

We are often asked what is the key difference between rule-based system vs. Machine Learning based system, in the context of insurance claims and waste, abuse and fraud (WAF) detection. We examined both approaches to see if Machine Learning is the real future of health insurance claims processing.

Rule based systems. Are they only as good as the humans creating them?

Rule based systems rely on two simple premises: set of facts followed by clear “if-then” set of rules. Perfect example of these two premises is the fact that men cannot get pregnant. This fact is then followed by “if-then” rule: if male patient claims healthcare insurance for pregnancy, then it will always be denied. Rule based systems are built by human experts with in-depth domain knowledge to guarantee best possible outputs. Hence, they are expert driven systems.

The biggest issues with rule based systems are that they are:

  • cumbersome;
  • unwieldy and require extensive human interference and constant updating;
  • poorly equipped to detect fraud, waste and abuse patterns in claims data.

In other words, rule based systems are slow to adapt, they are only as good as the human experts creating the rules and they need constant manual updating.

Machine Learning approach to claims analysis. Can Machine Learning based systems make their own judgement?

Machine learning approach examines large amounts of past data and focuses only on the outcomes from the experts. For example, a Machine Learning system is able to see that there are no male pregnancies in vast total number of claims and hence such claims should never be approved. No one needs to tell that to the Machine Learning based system as the software is able to make this logical deduction on its own by simply analyzing the data and looking for correlations.

Focusing on the outcomes rather than entire decision making process can make machine learning more flexible and less susceptible to some of the problems encountered with rules-based systems. The basic operation of a machine learning process is to say ‘given what we know about these historic outcomes, what can we say about future outcomes’. The more is the cumulative data, the more accurate the system becomes over time.

The key benefits of Machine Learning systems are:

  • they adapt by themselves and learn from the outcomes without any need for human interference;
  • Machine Learning models can use hundreds of inputs at the same time in order to give correct output most of the time.

And finally…

In the context of healthcare insurance claims this would mean multidimensional analysis of clinicians, specialties, primary and secondary diagnosis to treatments provided, medication prescribed and all of these compared to thousands of similar claims to detect possible signs of anomalies that are key indicator of possible fraud, waste and abuse. The resulting “taught” model is derived from the actual data itself rather than externally supplied expert information (as in rules based systems).

Would you like to learn  how our machine learning system works?

Contact us now!