We have seen Pattern Recognition in machine learning. Now, let’s look at some machine learning example to illustrate better.
Identifying Credit Card Fraud
Suppose you have credit card customer who is supplying their credit detail to Payment application in the grocery store. The challenge is to work out which of that transaction application should reject because they are likely to be fraudulent.
Like we discussed earlier in the pattern reorganization module, We can run historical data from the machine learning process to create a model. This model can be called by transaction application to make the decision.
Predicting Customer Churn
Consider little bit complex scenario. Suppose you have a problem of predicting customer churn. Imagine you have a mobile company and a lot of customers who call into the call centre. Call centre staff relay on some call centre application. Now you want to determine each calling customer is likely to be churn that is switching to your competitor from your call centre staff.
This has real value because you might offer a customer who is going to switch to your competitor. The call centre has huge calling data which is then needed to be aggregated using an aggregation application like Hadoop, Spark etc.
Now phone company combines aggregate data with other data like CRM to actually create data which can be used by a Machine learning process. It is common to use data from different sources as an input to the machine learning process. The result of this is Model which can be used by call centre application to get a good estimation to see whether the customer is going to churn. This also signifies Machine learning is commonly used with other data technologies like Hadoop or Spark etc.
Predicting Equipment Failure
Imagine you have a bunch of devices like robots or Thermostats which generate lots of streaming data which is handled by some real-time data processing software. This Software is looking for a certain pattern which can predict imminent failure.
If this software finds some failure then it contacts notification application which notifies the business user to take action like maintenance or fix. Well, the question is how real-time processing predicts failure.
We have historical data which contains information whenever the devices have failed. This data can be feed to machine learning as input to create a model which can be used by real-time data processing software.
A once again pragmatic solution to a very real problem that depends on good data and machine learning.