What does it mean to learn?
Until now, we understood why machine learning is very important. To understand this we need to know how did we learn to read? Well, it requires following things.
- Pattern Identification – For the reading purpose we first identify letters and then we look at the pattern of letters together to identify the word.
- Pattern Recognition – We have to recognize the pattern later when we see it again based on past experience.
This is how we learn to read. In a similar way, Machine learning works with data we provide to it.
Suppose We have a data of credit card transactions with only five records
For Credit card transactions we have 3 fields those are Customer Name, Amount and whether the customer is Fraudulent or not. From the above case, we can make out a fraudulent pattern it is obvious that the customer name starting with P are looks to be Fraudulent well probably not.
The difficulty of having so little data is that it is easy to find a pattern but difficult to find a pattern that is correct. In the sense, they are predictive and will assist us to understand whether the new transaction is expected to be fraudulent.
Now let’s consider we have more records of Credit Card Transactions
This time we have more records with more fields like the age of the customer, where the cards are issued and where it is used. If we closely look at it then we can easily find a pattern which is marked as fraudulent.
What is the pattern of fraudulent transactions?
For the above set of records, if the customer is his twenties and if the card issued in Australia and used in Russia and the amount is greater than $2000 then transaction is likely to be fraudulent.
Can we now confirm whether this pattern is predictive, Well probably not? We do not have adequate data. To do this well we need to have enough data then we can find the pattern or maybe use software that is where Machine learning comes in.
Machine Learning in a Nutshell
It starts with data that includes patterns then we supply that data into a Machine learning algorithm. The algorithm helps to detect a pattern in the data and then create a Model. A Model is a functionality typically code to identify a pattern when presented with new data.
An application can use that model by providing new data to recognize if this data matches the known pattern. For example, supplying data for the new transaction. The Model can return a possibility whether the transaction is fraudulent or not. Model discover this because it knows the pattern.