An Overview of Machine Learning Algorithms for Beginners

But first, what is Machine Learning?

What is the difference between AI and ML?

How do Machine Learning Algorithms work?

1. Collecting data from reliable sources
2. Cleaning data by removing unwanted/missing values, formatting them and splitting them into test data and training data
3. Choosing a Machine Learning model (explained further below)
4. Training the model by analysing the patterns and making predictions based on training data (from Step 2)
5. Assessing the Machine Learning model with test data (from Step 2)
6. Tuning the model parameters to improve its accuracy
7. Using the model on unseen data

What are the 3 main types of ML models?

1. Supervised Learning
   As the name suggests, supervised learning is when a Machine Learning algorithm or a Machine Learning model learns with the help of a supervisor. This means, there is a feedback system that explains the model whether or not it is working correctly. Let us understand this with an example. One of the most basic ML algorithms is the image classification model, which distinguishes between whether an image is that of a cat or a dog. If the model guesses correctly, then the supervising entity has nothing to do. But if the model guesses incorrectly, then the Machine Learning Engineer has to tweak the parameters so that it works properly. A more complex example is to do sentiment analysis on a piece of text like a user’s tweets. Such a model will try to understand whether the user (or a customer) is happy with an experience. Supervised learning algorithms are further divided into supervised classification algorithms and regression algorithms.  We will explore ML classification algorithm and regression algorithm in depth in our subsequent blogs.
2. Unsupervised Learning
   In the above examples, the data is labelled. In unsupervised learning, the input data for a model is unlabelled and the model has to recognise patterns in the same. For example, the demographic of customers that is likely to buy a particular product. A store may not always collect all the inputs, like age group, but based on other purchases, the model has to make an estimate and classify the user accordingly.
3. Reinforcement Learning
   In a reinforcement learning model, the machine/agent to begin with understands two things - positive feedback and negative feedback. Then the agent interacts with the environment and checks the kind of feedback it has received and makes adjustments. A simple example for Reinforcement Learning would be a product recommendation system, where the model recommends a product to a customer and based on the feedback from the customer fine-tunes its recommendations.

What are some Machine Learning applications?

1. Image recognition
   One of the most well-known uses of Machine Learning is in social media applications which suggests users which friends to tag in which photos. Social media also makes use of other aspects of Machine Learning to suggest pages and accounts to follow and more!
2. Speech recognition
   With the advent of Siri, Alexa, Cortana and Google Assistant, speech recognition has become an everyday part of our lives. Here, these virtual assistants use Machine Learning models to follow commands based on voice instructions.
3. Traffic prediction
   Considering how rapidly the infrastructure of cities is changing, traffic prediction has become one of the essential applications of Machine Learning. By analysing the patterns of traffic on a daily basis, systems (like Google Maps) are able to accurately predict traffic at any given point on a route.
4. Email & Spam filtering
   This application of Machine Learning is something you must have already observed. The filters in our email can mark mails as important/not important/promotional/social/spam and even blacklists.
5. Medical Diagnosis
   Machine Learning can also be used for diagnosing diseases, including charting the position of lesions in the brain. In fact, one of our projects in the Artificial Intelligence & Machine Learning certification course is on melanoma detection.

Conclusion