What is Machine Learning and What are its types?

Hey Guys, Today in this article, we are going to look at the Types of Machine Learning very deeply! So, much deeper, that this topic would be clear to you like water.

Machine Learning

We would start from scratch and would go to the deep level!

I could betcha, you would understand every word of this article! Just read till the end!

So, let’s start this beautiful article, “Types of Machine Learning” without waiting for a second.

What is Machine Learning?

So, as I said earlier, We will start from scratch so, that everything will become clear for you!

Let’s start with the starting!

Around the 1950s, the term “**Artificial Intelligence**” was first introduced to the world!

This term means, nothing, but transferring human intelligence into machines, so that, they would be efficient as as humans!

But **Intelligence is not a small thing. It refers to Emotional Understanding, Pattern Recognition, IQ, Problem-Solving Skills, Critical Thinking** and a lot more!

But, we can’t develop so many skills in machines, and thus we focus on one thing in the initial days.

So, in 1955, Symbolic AI was created, in which a huge knowledge base was created using the if/else conditions! A lot of if/else conditions.

then, Expert Systems, came into existence, in which again, through a lot of if/else conditions, a system was created that could focus on a particular area!

However, this concept of Expert Systems has a problem. Think for yourself, suppose you are writing a code for identifying dogs in Python!

So, what do you think how many lines of code would it be?

It could be infinite! Because, there are a lot of dogs bred, a lot of characteristics. How much could you cover to identify a dog?

Thus, the solution to this main problem is Machine Learning.

In booking language, It is a field of Computer Science that uses statistical techniques to give computer systems the ability to ‘learn’ themselves with data without being explicitly programmed.

What is Machine Learning?

This definition is much easier to understand😂! Kidding! Let me explain it to you!

How does ML work?

Like in Expert Systems, we have written programs for every condition, but in ML this is not the case.

We don’t have to code! We just have to provide the data, in which there would be input and output, to the machine!

And what it would do? By Using Machine Learning Algorithms, it would draw out the patterns, between the input and output!

And then, if you provide the input to that machine, it will provide you the output by itself!

I hope you understand it!

Types of Machine Learning

Now, we have reached into that part of the article, for which you all are here, the Three Types of Machine Learning.

All those three types are listed below -:

• Supervised Learning
• Un-supervised
• Semi-supervised Learning
• Reinforcement Learning

They are deeply described below.

Supervised Learning

Supervised learning, could be understood as simple Machine learning!

This means, in Supervised Machine Learning, there are both input and output cells, the machine finds the pattern in it and then predicts the output for new data!

As simple as that!

Example -: you have a data of 5000 students! The data is about their IQ, CGPA and their Placement (Y/N)

Here, the IQ and CGPA are input cells, and Placement is the output data cells.

The supervised learning is divided into two main parts -:

These are described below!

But before, Regression and Classification, let me tell you something about the type of data!

There are two types of data, These are -:

• Numerical Data -: Numerical data means, the data in the form of numbers. Example — age, Marks, IQ, CGPA etc.
• Categorical Data -: Categorical Data means, the data in the form of English characters. Example — Placement, Breed of the Dog, Name of the person etc.

Regression

So, now regression means when the data in the output cells is in the form of Numbers (means, Numerical Data)

Classification

Classification, means, that when the data in the output cells is in the form of characters (means, categorical data) it is known as Classification.

Advantages of Supervised Learning

• Easy Decision-Making: Supervised learning makes decision-making easy by learning from labelled data. It’s like teaching a computer to recognize patterns based on examples, making it great for tasks like image recognition, where it learns to identify objects from labelled images.
• Accurate Predictions: It excels at making accurate predictions by learning from known outcomes. This is like a smart assistant that learns from past experiences to predict future events, making it valuable in fields such as finance for forecasting trends.
• Adaptability: Supervised learning adapts well to new data, adjusting its understanding based on fresh information. It’s akin to a learner who gets better with more practice examples, allowing it to stay relevant and effective in dynamic environments.
• Quick Problem Solving: It is proficient in solving problems quickly by leveraging past knowledge. Similar to a skilled worker who uses previous experiences to tackle new challenges efficiently, supervised learning excels in tasks where it can apply learned patterns to solve problems.
• User-Friendly Applications: Supervised learning is user-friendly, making it accessible for various applications. Just like user-friendly software, it doesn’t require extensive programming expertise, allowing users to implement it in diverse areas like healthcare for disease prediction, where accurate predictions are crucial.

Disadvantages of Supervised Learning

• Need for Labeled Data: One drawback is the requirement for labelled data, where each example needs to be labelled with the correct answer. This can be time-consuming and expensive, especially in cases where obtaining labelled data is challenging.
• Limited Generalization: Supervised learning may struggle when faced with new, unseen data that differs significantly from the training set. It’s like a student who’s only good at solving specific types of problems but struggles with unfamiliar ones.
• Overfitting Concerns: There’s a risk of overfitting, where the model becomes too specific to the training data and performs poorly on new data. It’s akin to memorizing answers for a test but not understanding the underlying concepts.
• Sensitive to Outliers: Supervised learning can be sensitive to outliers, which are unusual data points that deviate from the norm. This is similar to a grading system that gets heavily influenced by extreme scores, impacting the overall accuracy of the model.
• Dependency on Quality Data: The performance of supervised learning heavily depends on the quality and representativeness of the training data. If the data is biased or incomplete, the model’s predictions may also be biased or inaccurate, resembling a teacher trying to instruct students with flawed materials.

Unsupervised Learning

Now, comes the Un-supervised Machine Learning! So, what is unsupervised learning?

Let’s discuss it out!

In Unsupervised learning, we have only input data cells and no output data cells!

Example -: You have data of 2000 students, about their class and school name!

This data is an example of Unsupervised Machine Learning, as it has only Input Data Cells (Studying Class and School Name), but no output cells.

It is divided into four main types, which are -:

• Clustering
• Dimensionality Reduction
• Anomaly Detection
• Associate Rule Based Learning

All these are briefly described below.

Clustering

It is one of the most important concepts of Unsupervised Learning!

In most of the problems, you are going to use Clustering Algorithm to solve it. I Betcha!

So, what does it mean? It means, converting a large amount of similar data into many clusters.

This helps in processing and analysing the data, as it is much shorter now, because of the Clustering algorithm.

Example -: You want to know how much area, a home will become, and for this, you have collected the number of rooms, number of kitchens, and number of bathrooms.

You collected this data from 100 people.

Now, what could we do, we could substitute all these input data cells with only one data!

Which could be the simple Square fit area, which is the cluster of a Total number of rooms, kitchens and bathrooms!

This example perfectly describes the Clustering Algorithm.

Dimensionality Reduction

Now, comes the dimensionality reduction!

It means, removing unnecessary input data cells, those data cells that don’t affect the output, and removing garbage, is Dimensionality Reduction.

Example -: You have data of 3000 students studying in class 10, about their study duration, the books they read and their skin colour!

And you also have whether they score more than 90% in class 10th.

Now, in the data, you can see, that the input data of skin colour doesn’t have any relation with the percentage score!

So, what is the meaning of having this input data cell? The Dimensionality Reduction will reduce it!

Anomaly Detection

Anomaly Detection means, something different or shocking.

Let me explain it to you by an example.

Suppose, our Machine Learning Model detects that a student who studies less than 3 hours, couldn’t score more than 90% in class 10th.

But, a student has done it!

So, it is an outlier, and anomaly detection removes it from the data.

Associate Rule Based Learning

Now, comes the Associate Rule Based Learning.

It refers to a type of analysis that aims to discover interesting relationships or associations among variables in a large dataset.

This means, analysing the patterns between the input and output of large data sets.

The most well-known algorithm for association rule-based learning is the Apriori algorithm.

Semi-supervised Learning

We have reached the third type of Machine Learning, which is Semi-supervised learning.

As the name himself suggests, it is a semi-supervised. This means, mixing up both supervised and unsupervised learning.

I am sure you won’t use Semi-supervised too much, it would have only some limited uses.

So, now let’s directly discuss its advantages and disadvantages.

Advantages of Semi-supervised Learning

• Efficient Use of Data: Semi-supervised learning efficiently utilizes both labelled and unlabeled data, maximizing the benefits of available information. It’s like having a mix of solved and unsolved problems, enabling the model to learn from both to improve its overall understanding.
• Cost-Effective Labeling: Semi-supervised learning reduces the cost of labelling data since it can leverage a smaller set of labelled examples. This is akin to having a teacher mark only a few key questions in an assignment, saving time and resources.
• Improved Generalization: It often leads to better generalization compared to purely supervised methods. By learning from a combination of labelled and unlabeled data, the model becomes more adaptable to diverse scenarios, similar to a student gaining a deeper understanding by solving a variety of problems.
• Flexibility in Real-world Scenarios: Semi-supervised learning is flexible and applicable in real-world scenarios where obtaining labelled data is challenging. This is like a student who can learn from both guided lessons and independent study, adapting to different learning environments.
• Enhanced Model Accuracy: Incorporating unlabeled data aids in refining the model’s accuracy, particularly when faced with complex tasks. It’s like a learner who gains a more comprehensive understanding by exploring additional resources, leading to improved performance in various applications, from image recognition to natural language processing.

Disadvantages of Semi-supervised Learning

• Limited Guidance: One drawback is the limited guidance from labelled data, which may hinder the model’s performance in tasks requiring precise categorization. It’s akin to a student having only a few correct answers but needing more guidance for a comprehensive understanding.
• Potential Noise Impact: Unlabeled data may contain noise or irrelevant information, impacting the model’s accuracy. This is similar to having background chatter while trying to focus on important information, potentially leading to confusion and misinterpretation.
• Challenge in Unlabeled Data Quality: The quality of unlabeled data can vary, and its usefulness depends on the relevance to the task. It’s like having study materials that may or may not align with the curriculum, making it challenging for the model to extract meaningful insights.
• Complex Model Tuning: Semi-supervised learning models may require complex tuning to balance the contributions of labelled and unlabeled data. This is similar to adjusting the difficulty level of questions in an exam, and finding the right balance can be intricate.
• Sensitivity to Initial Conditions: The effectiveness of semi-supervised learning can be sensitive to the initial conditions and the choice of parameters. Similar to a student’s performance being influenced by the starting point of their studies, the model’s outcome may be affected by its initial setup.

Reinforcement Learning

Now, only the Reinforcement Learning has left!

Believe me, this type of ML is super, which means, I can’t express my feelings! Let me tell it to you quickly!

See, in Supervised Learning, we have both input and output data cells, but in unsupervised learning, we have only input data cells. Okay!

But, but, in reinforcement learning, we have no data, neither input nor output.😂😂

Seriously! You could check it out on Google. It learns everything by itself, by living in the environment surrounding it.

As humans learn!

Self-driving cars are the best use case of Reinforcement learning. It is a new domain in Machine Learning, but slowly it is establishing its root.

Now, let’s discuss some of its advantages and disadvantages.

Advantages of Reinforcement

• Learning by Doing: Reinforcement learning is like learning by doing. The system gets better at tasks through trial and error, much like a person learning to ride a bike by practising.
• Adaptability: It adapts well to changing environments. Imagine a robot learning to navigate a room — it can adjust its strategy based on the layout, much like a person finding the best path in an unfamiliar place.
• Autonomous Decision-Making: Reinforcement learning enables autonomous decision-making. It’s like a smart assistant making decisions on its own, learning from experience to perform tasks without constant supervision.
• Versatility: It’s versatile and can be applied to various tasks. Whether it’s playing games, controlling robots, or managing resources, reinforcement learning is like a multi-talented learner that can handle different challenges.
• Continuous Improvement: The system keeps improving with more interactions. Similar to a person getting better at a game with practice, reinforcement learning continually enhances its performance as it interacts more with its environment.

Disadvantages of Reinforcement

• High Computational Requirements: Reinforcement learning often requires a lot of computing power. It’s like needing a powerful computer to handle complex tasks, making it less practical for certain applications with limited resources.
• Exploration Challenges: The system may struggle with exploration. Imagine playing a new game without knowing the rules — the learning process may take longer, and the system may not discover the best strategies easily.
• Delayed Rewards: Reinforcement learning relies on delayed rewards. It’s like waiting for the results of an action in the future, which can be challenging for the system to connect the consequences to specific actions.
• Sensitivity to Hyperparameters: It’s sensitive to hyperparameters. Adjusting the parameters is similar to finding the right settings for a machine, and getting them right can be tricky, impacting the effectiveness of the learning process.
• Limited Interpretability: The decision-making process can be less interpretable. Understanding why the system made a specific choice is like deciphering a complex decision-making process, which may limit its application in certain fields where transparency is crucial.

Conclusion

So, guys, finally we have completed the article whose title is “Types of Machine Learning”.

I hope you understood every concept mentioned in this article! For any query or any assistance, you could comment to me!

I will make sure to reply!

So, This is Dipansu Joshi signing off for today with a smile. Meet you in the next such discussion.

Till then, Keep Learning! Keep Exploring! Ba-Bye!

Originally published at https://dipansutech.com on December 22, 2023.