INTRODUCTION TO MACHINE LEARNING
People always ask what machine learning is, can a machine learn and how is that possible? Guess what Machines can learn and we can teach them to learn and some can even learn own their own. Through learning, machines are now able to predict some events of the future with the help of past events. Interesting right?
This article is for beginners who are interested in gaining knowledge about Machine learning and how machines learn.
When some people hear about machine learning, they eventually think of robots and all. Well, machine learning is an Artificial Intelligence application that enables self-learning of a machine to do certain tasks on its own without human intervention. It achieves doing this by using old data to determine an event when new data is feed into the machine.
Machine learning is one of the most interesting concept when it comes to Computer studies. It does not only cover computer studies but it’s an all-round concept. The good thing about it is, it can be applied in numerous areas in our daily lives such as:
Health – Predict and diagnose a patient.
Finance – Predict and detect fraud.
E-commerce – Recommendations and predict customer churn.
Biology – getting patterns in gene mutation that shows signs of cancer or other disease.
You see, Machine Learning can be applied anywhere as long as there is data. Now let’s take a look at how machines learn.
How do machines learn?
Good question there, how do machines learn? Well this is how they learn. Machine Learning is all about building mathematical models to get an understanding of data. Data is acquired and it is feed into the machine. The machine learns to group the data according to its similarity or by gaining patterns that are interesting. But then not all the data collected is good, we do get bad data. We try by all means to separate bad data and good data. Imagine driving to some new grocery store in town and ask for directions, if you get good directions you get to the grocery store fast. But if you get bad directions you totally get lost and if you get good and bad directions, you get lost and eventually find your way but after a long struggle. We don’t want to get lost and struggle in finding our way. That’s why we need to come with an unbiased model.
So the Important thing about machine learning is feeding the machine with good data to get accurate and better results. Machines learn by getting old data being feed to it and predict about what’s likely to happen in the future. It’s like getting past events to determine the next move on new events. For example if a child touches a burning candle and gets burnt the child has learnt, and if later on they see a burning candle they don’t touch it cause they know it burns.
There are different types of machine learning and ways a machine can learn, that are:
Machine learns from labelled data. Meaning the data has a guideline that shows it that is a right or wrong answer. It classifies the right answers from the wrong answers. For example training a machine to identify spam emails. You feed the machine with spam and no spam emails and when the machine is feed with new data, its able to identify spam emails with the help of the past data and algorithms. Another way is being able to predict the price of a car given details such as mileage, brand etc. Supervised learning is divided into Classification and Regression. Classification involves predicting a label/categorical variables such as Yes/No whereas Regression involves predicting quantity such as weight/height. In Classification, labels are in discrete form whereas in Regression labels are in continuous form.
Algorithms used in supervised are:
• K-Nearest Neighbours
• Linear Regression
• Logistic Regression
• Support Vector Machines (SVMs)
• Decision Trees and Random Forests
• Neural networks
It involves the machine learning from unlabelled data. This means that the machine does not have a guideline to show it what is right or wrong. It has to come up with patterns from the data to make up a right or wrong answer and group the patterns. This is where a machine tries to learn own its own without human intervention. Unsupervised learning is divided into two categories, which are Clustering and Association. In clustering algorithms identify distinct groups of data. Association focuses on a rule like an individual’s purchase buying pattern and groups that together.
Algorithms used in Unsupervised are:
• One-class SVM
• Principal Component Analysis (PCA)
This type of learning uses both labelled and unlabelled training data. It’s a combination of supervised and unsupervised learning algorithms. Semi-supervised is useful when incomplete labels are available in the training data.
This type of learning is a bit different from all the other techniques. It involves an agent, this agent acts as the learning system. The agent learns from its environment by observing its surroundings, then it performs an actions based on its environment. When it performs an action, it gets a reward in return or gets a penalty as a negative reward. The agent must learn until it reaches its goal and choose the best strategy for learning, known as a policy. By reaching its goal it attains the most reward over time. A policy is defined as the action the agent takes when it is in a given situation.
Well that’s a wrap, we learnt about what Machine learning is, and how machines learn. I bet from this you now want to learn more and dig your hands in the dirty work and teach your machine one or two tricks. Check out the next blog post and enjoy the learning curve.