Machine Learning 101
In the previous part — Part 2: Data Patterns, Types of Variables, and Algorithms we have gone through various types of Data Distribution that are commonly used in Machine Learning, we learned what are Variables and their types, and finally different types of Machine Learning Algorithms. In this part, we are going to dive deep into one of the interesting concepts Types of Machine Learning Models, and when they are used or applied.
We know from Part I: Introduction, that a machine learning “model” represents what was learned by the machine learning algorithm from analyzing the data.
The different types of Machine Learning Models are —
1) Supervised Machine Learning
Let’s suppose there is a security camera that automatically identifies or recognizes the face of an employee or student standing at the gate of the college/office premises. If registered, allows them to enter the premises.
In this case, the Machine Learning model will learn from several features of a human face, such as the shape of the eyes, nose, mouth, color of the eyes, etc, as well as the corresponding name (target) associated with the face. When a new image is captured using the camera, it takes the face features from that image as input and predicts the name of the student or employee as the target/output if registered.
Note: The data is labeled in Supervised Machine Learning.
Supervised Machine Learning can be further categorized as —
- Regression — Model that predicts a numeric quantity, such as house prices, ice cream sales, etc. Example algorithms: Linear Regression, SGDRegressor, etc.
- Classification — Model that predicts a category, such as whether the patient is infected by covid or not, whether the customer will be eligible to pay a loan or not, etc. Example algorithms: Decision Tree, Support Vector Machine, etc.
2) Unsupervised Machine Learning
We may have seen that an offer or discount on products is provided only to some people and not to others. Let us understand the mechanism behind it. For example, an electronics brand has information about its customers such as age, gender, occupation, education, personal preferences for various products, etc.
Using this information the machine learning algorithm finds some similarities and differences between the customers, and groups them. Customers in each group can then receive personalized offers and discounts.
Note: The data is unlabeled in Unsupervised Machine Learning. There is no specific target associated with the features.
Unsupervised Machine Learning can be further categorized as —
- Clustering — Model that groups data based on similarities, such as Customer segmentation. Example algorithms: K-Means Clustering, Hierarchical Clustering, etc.
- Association — Model that finds dependencies between data, such as Market Basket analysis which groups products that can be purchased together. Example algorithms: Apriori, FP-growth, etc.
3) Semi-supervised Machine Learning
Let us continue with the customer example, to better understand Semi-supervised Machine Learning. We know that using the Unsupervised Machine Learning algorithm, the customers are grouped based on their similarities and differences. Let’s suppose, some of the customers have also specified their preferred type of appliance such as office, home, and beauty appliances, and the electronics brand wishes to predict the same for the new customers.
In this case, the information regarding the preferred type of appliances is incomplete for some customers and complete for others. Also, we cannot use only one of either Supervised or Unsupervised Machine Learning because we have a mixture of labeled and unlabeled data.
In such a scenario, Semi-supervised Machine Learning plays a great role where it first uses Unsupervised Machine Learning to group the customers, we can then label those groups (office, home, and beauty appliance), then Supervised Machine Learning is used on the labeled data to predict the preferred type of appliance for the customers.
Note: Semi-supervised Machine Learning combines Supervised and Unsupervised Machine Learning. In other words, act as a bridge between them and overcome their flaws.
4) Reinforcement Learning
To understand Reinforcement Learning, let us take an example of a video game. In Reinforcement Learning, each player in the video game acts as an agent, the game itself is called an environment, the move of the player at each step is called a state, and the goal is to obtain the highest possible score. Here, when we take a right step or action, we are rewarded and when we take a wrong step or action, we lose points.
We will deep dive into Reinforcement Learning in the upcoming parts.
Stay tuned, in the next part we will understand What is Linear Regression? and How it works? Please share your views, thoughts, and comments below. Feel free to ask any queries.
References: