Introduction to Machine Learning
Machine Learning is the latest buzz word that has caught world’s imagination and in this article we will give a very intuitive introduction to machine learning for beginner. Till 4-5 years back machine learning was mostly confined within four walls of educational & research institutes and was used industrially only by few handful of giants like Amazon, Google, Facebook to it’s full potential.
But the machine learning landscape has changed drastically in last 3 years. Suddenly there is a boom in industrial and main stream usage of machine learning and AI. Companies – small and big alike, across all domains are now trying to implement machine learning in their core functional areas. With this sudden boom – the demand of machine learning skill and lucrative salary prospect or fear of job loss has created a stir in the job market. People who probably never heard about the word “Machine Learning” till few years back are now suddenly running from pillar to post to understand what machine learning is really all about and if they can re-skill themselves with the change in market demands.
Since available Machine Learning resources are mostly mathematical or complex, it causes a scare within beginners. So let us try to understand machine learning.
Traditional System Vs Machine Learning System
The best way to under machine learning is to compare it with the traditional approach of building system
Traditional System
The traditional computer systems that are getting designed for many decades are explicitly programmed to take logical steps or decisions based on input data.
Here a subject matter analyst understands the requirement and designs an algorithm for input data. The algorithm is programmed in the Traditional System. Now the input data is run on the Traditional System and the output is obtained.
Machine Learning System
In machine learning system, the algorithm is neither designed nor programmed explicitly. The beauty of such system is that it derives the logic based on some past sample data.
Here past sample data is run on Machine Learning system and an algorithm is generated based on it. Now any new input data can be run on this algorithm to produce outputs.
An experienced person in this field might notice that it is not an accurate representation of machine learning systems but this is good enough to build an intuitive foundation of machine learning for new beginners.
Case Study – Loan Approval System
To take our intuitive idea one step forward let us consider a sample case study.
Let us consider a small bank that is planning to build a Loan Approval System. Such systems go through the details of loan applicant and decides whether to reject the loan application or approve it (by making a prediction that the applicant might or might not default on the loan). We will see how we can make this system using both traditional and machine learning approaches and how both these approaches stand against each other.
Traditional Approach
In this approach, bank employs an analyst who is having domain knowledge of how finance institutes and loan works. Analyst studies the details of past loan applicants with the bank and their loan repayment history. Analyst then tries to identify the list of applicant’s details that were mostly common within the defaulters. Based on his studies and findings analyst designs a set of rules and filters in the Loan Approval System to accept or reject loan application.
Machine Learning Approach
In this approach, the past details of past loan applicants with the bank and their loan repayment history is fed to the machine learning system. The system goes through the past data, understands some hidden trend and pattern in the data and then generates an algorithm based on its observation of past data.
In machine learning field, feeding of data is more formally known as training and the algorithm generated after training has formal term of machine learning model.
This machine learning model generated, then becomes the main brain of the Loan Application System that decides whether to approve the new loan applications or to reject it.
Comparison between both approaches
As you would have realized by now, traditional approach are having some serious short comings and machine learning is winning this fight between man vs machine.
- The design of traditional system is completely dependent on how analyst perceives the past data.
- It is not humanly possible for any analyst to understand the most underlying and hidden information within the data.
- If number of details of loan applicants are high it becomes increasingly difficult for analyst to come up with good rules and filters with manual analysis even with best of tools
- Machine Learning system on other hand is capable of understanding hidden information even if number of details of loan applicants are high.
- If new demographic is introduced in the loan application, the rule based traditional system will be very difficult to scale up, whereas the machine learning system can be scaled very easily with new type of data.
- As the bank grows its business, the traditional system might also grow on to become a big chaos of rules and filters that is very difficult to maintain and debug.
- On other hand the machine learning system can still be scaled easily with growing complexity of bank’s business and it will still remain a simple system.
We can go on to carry more in depth analysis between the two type of systems, but I guess these are good points to intuitively understand why machine learning system is a better choice here.
Machine Learning – Definition
Now that we had a small glimpse of Machine Learning, I think we are ready to give a definition to it.
Machine Learning does not have one formal definition. Many scholars and researchers have come up with their own definition of machine learning ranging from complex ones to simple ones and they are all correct.
I leave you with the following definition by Arthur Lee Samuel, pioneer of this field who defined Machine Learning as –
“Field of study that gives computers the ability to learn without being explicitly programmed.”
