Machine Learning for Beginners | Full Guide | Webinar

Key Takeaways

So hello everyone, so first of all welcome to the webinar on Smart Machines Smart World Beginner's Guide to Machine Learning. Okay, so this is the topic we are going to discuss today . So before going to our topic, let's take a look at a few things about our app. Okay, so Entry App is one of the best learning apps in India and a secure gateway to jobs. Okay, and Entry was launched in 2017. We have 6+ courses available in all Indian languages. Also, 10 million+ downloads are available on the store. Also, we have 10,000+ placements in government and private sectors. Okay, our mission is to provide quality content to our users, that is, our learners. Also, our mission is to make them employment-ready. Also, the learning materials we provide here are in the mother tongue of the users, so they can learn very easily. Okay, and Our vision is to become one of the best job securing apps in India. To know more about our vision and entry, our courses, mentors, student success programs, and our placement partners, just use the link below. My name is Greeshma. I have six plus years of training experience and have trained 5 plus students. I also specialize in machine learning technologies, data analytics, and data science. Okay, let's start first. What is machine learning? Okay, machine learning. Before we actually go into machine learning, one thing we need to know is that AI stands for Artificial Intelligence. Because it is a branch of Artificial Intelligence, machine learning is called AI. Okay, so wherever we look, AI is everywhere. For the past week, our feature called Nano Banana has been trending. If we look at anyone's status, whether it's on WhatsApp, Instagram, or Facebook, we see a lot of posts of images created using Nano Banana. So what are these created using AI? When we ask people about it, they say, "These are AI-created images created using AI." Exactly, when we say AI, what is A? We all know what artificial intelligence is in full form. Okay, so when we say artificial intelligence, we mean making machines think the way humans think or making machines do what humans can do. So, what do we mean by artificial intelligence? Branches are coming, one of them is machine learning. Not only machine learning, there are many other branches like computer vision, robotics, natural language processing, and so on. What do you mean by branches? Our artificial intelligence is called because artificial intelligence is such a broad area that we can see it every week, week by week, new updates are coming in whatever it is. Now, for the last few weeks, it has been said that this is one of them. Okay, so that's what AI is called. Machine learning is a branch of AI. Okay, come on, let's talk about the topic. What did we say? Machine learning is a branch of AI. So what is machine learning? When we say machine learning, we mean computer learning from date and time to performance programming. Instruction means nothing. If we put it very simply, that's all. We will teach the machines with a lot of data. Okay, we will teach them with a lot of data. So what do they do after studying and learning? They must have learned and learned a lot of data. They must have had a kind of experience. Okay, then from that experience, they will understand what results to give to the data that comes next. That is, they learn a lot of things using the data we give them. In the end, they will produce the results we want without us having to say anything. We don't have to give anything explicitly. There is a capability to give results. This technology called machine learning has that. That is, machine learning is the best example of this. When we go to school, when we send students, what do we do with the notes that teachers give us? When we study and study, don't we? Okay, then what do we get after studying and learning those notes? A kind of experience is when teachers ask us a question, what do we understand? After understanding what it is, we will first have the ability to give a quick answer. It is the same in this machine. Okay, if we take the case of Max, that is a great example. How do questions come in Max? We first solve the particular problem that has been given to us, we completely understand how to solve it. Then when a question comes for the exam, we use all the techniques we have learned to solve it. This is the same with our machines. Okay, what can we do here? We can't do it by heart. We have to understand everything and come to understand it. How do we learn by practice? We can only do it by doing it. Otherwise, we will not be able to understand it properly after just beating it by heart and going to the exam. Okay, machine learning is like this. Okay, now when we say traditional programming, what is traditional programming? We usually use C++ and Python. When we say normal programming, for example, to add two numbers, what do we need to add two numbers? In the case of programming, we need two variables. Then we use two variables to store two numbers. To add two numbers, we need two variables. In the number called A, I store the value called 10. In the variable called B, I also store the value called five. Now, let's add this. Our requirement is that what should we give to add? We need to give a formula. So what is the formula that we give there? Ceq. B. This is a rule that we usually give using traditional programming. What is traditional programming? First we give some data, first we give our program, after giving the program, we give the data, what is our data here, what is it? Five is our data, then we give those two numbers to conventional programming, what is conventional programming, what is the problem that we solve using our traditional programming language, we have to add two numbers, then we give the data and what do we get from it, the result. Now, let's take the case of machine learning, first what do we give here, we give the data, after giving the data, along with the data, the machine gives us something else, we give the result, and when we give that result, what do we get from there as a generated program? Here, when we say a program in the case of machine learning, actually it is a model, machine learning can be developed for us, using a technique called machine learning, we can use it to develop a program. When we provide the results, we create a model using all the machine learning algorithms . Okay, we will then make predictions using that model. This is what happens if it is on the machine . Okay, again, if we're saying it very simply, we're saying to a small child that they've never seen an object before, and now that child has never seen an animal in their life that they call a cat. Okay, if that's the case, when the child sees a cat for the first time, we tell the child that this is a pooja. The next day, let's imagine that the child sees another kind of cat. There are many varieties of cats in pooja. So imagine that the child sees a different type of cat. So when the child sees another type of cat, he can't say that it is a cat. Now, for a few days, we show the child images of different types of cats and tell the child that this is a cat. This is a pooja. This is a pooja. Then, day by day, when we keep telling the child like this, what will the child understand about the features of that cat? What are the features? How does the eyes sit? How does the nose sit? And so on. What will the child do with each feature? And so on. When you say, "What is a cat?" Because the child has understood what the features of a cat are, what are its four legs, what is its nose, what is its eyes, and all that. The child has completely understood and taken all that. The child has put the child's experience and now we have to understand and take that particular image that was shown to us as a cat's. Okay, this is what happens in machine learning. Okay, now let's use a real example of machine learning. Now, when we go to the bank, we apply for a loan. When we apply for a loan, we use machines to predict whether the loan we applied for will be sanctioned or not. So, how can a machine do that? Our bank has not only a lot of transactions, but also a lot of loan application dates. When we say the date of the loan application, they will have various types of data in their hands. This is a particular example of ours. Has the applicant taken a loan before? If so, how much has been taken? What is their credit score? Has the loan been repaid? Many such features. That particular data. Then the machine uses this data to understand everything. It understands the patterns in that data. Then, after putting all this data, the machine will have an idea. So, when someone comes, it says this and that. When a person with these features comes, the system will understand whether he should pass or reject this particular loan. Yes or no. Okay, so what does the system do? It makes predictions. Okay, so this is one of the examples. It is said that it is a case of machine learning. Okay, now in the case of machine learning, the most important thing to mention is that there are three different types of machine learning techniques. We have said what machine learning is. Similarly, there are many categories in machine learning. Mainly, we There are three types of learning: supervised learning, unsupervised learning, reinforcement learning. The first one is supervised learning. When we say supervised learning, we mean training data to predict outcome on unseen examples. That is, when we give a machine a piece of data, we also tell the machine what kind of data it is. Supervised learning is such a learning technique. As its name suggests, there is someone there to supervise. It's about having someone there to teach. For example, let's say I'm giving the data of an email. When I give the data of an email, I'm saying that the email that came in is spam. Okay, so when I say spam, what does that particular system understand that the email that came in is spam? I'm giving the next data for that, what I've given is not spam. Here, spam and not spam are our classifications. That is, when we take the EA email spam detection or when we consider an application like that, is the email that came in spam or not spam? When we open our mails, there is a folder on the side of our Gmail and on the side that says spam. So, if the incoming mail is spam, it automatically goes to that folder. How does it go? When an email comes to our mail, its email filter adds it. What does the email filter do? If the incoming mail is not spam, it also sends it to the inbox. If it is, it is not classified as spam, it is a kind of classification. This is what we call supervised learning. That is, in this technique called supervised learning, when we say machine learning, what is it? It is completely data-dependent. Only if we give the data can the machine learn what to do. So when we give the data, we are telling the machine that this incoming data or incoming mail is spam. Or we are telling the machine that this email is spam. So, what does it do, after all this data that we have all given, the machine will develop a model . Okay, as we saw earlier, as we saw in the previous figure, what is a program that develops? Here, in the case of machine learning, what is a program? What is our model? We will develop a model. Using that developed model, we will make predictions. That is, we will put the data that we have in our hands for so long and study everything. Now, what comes next is what we are testing, right? Just like we test for exams. Okay, so when we do this, if there is data coming from outside, our system will predict whether that data is spam or not. Okay, so this is what we call supervised learning. Now, if we say the case of medical diagnosis, is it cancerous or non-cancerous, we can classify it. Similarly, if it is the case of price forecasting, how much will the stock price be in the future? We will be able to predict each of those things here. In this way, we will give data and teach it. That is what we are supervising. What is called learning now is unsupervised learning, unsupervised learning, supervised learning, the exact opposite. Here we only give data, we don't give labels for it, we give answers for it. We just give data. Okay, we give only data and the machine automatically finds out what the answers should be from this data. So, we call such a technique unsupervised learning. That is, I have a big bag in my hand. Okay, imagine that there are three colors of balls in a big bag. There are some red balls, there are some yellow balls, there are some green balls. So what are we doing here in the case of unsupervised learning? We take each color of balls into groups. Okay, we take only green balls as a group, only red balls as a group, and only yellow balls as a group. So we put each one here as a group. Okay, so we don't give answers, we give only data, we find hidden patterns in the data, and we use that pattern to divide each one into groups. What we call unsupervised learning is what we see here, what we see in this figure, inside this circle, what are the customers? So what are the customers? We divide customers into groups based on their preferences and interests. Okay, that's what we call customer segmentation. So customer segmentation is one of the examples. So, when we say customer segmentation, as I said, we divide users into groups based on their age, their interests, their spending habits, or their location. For example, we watch a lot of videos on Netflix, in other words, we watch a lot of movies and shows. If we watch a movie on Netflix, we only get suggestions for action movies. So there will be many customers who only like action. Then, we divide all such customers into a group and then divide the customers who only like romance into a group . Okay, then we'll divide those who watch things like cartoons into another group. So that's what customer segmentation actually means. A company understands its customers . After understanding their customers, they can share personalized content with them. Okay, so in a way, this is used to help businesses. Okay, so we can provide users with personalized offers, ads, and products. Okay, what happens when you do that is that it helps business . Okay, so that's what we mean by customer segmentation here. If we say another application or an application in unsupervised learning, anomaly detection is anomaly detection. When we say anomaly detection, if we see a pattern that is unusual, what can we say that it is an anomaly? Okay, if we say that one of our cards was used to spend one lakh rupees and that too in another country, what does that mean? That is actually a fraud, right? Anomaly detection is detecting such things. When we say anomaly, we can use this machine learning technique called unsupervised learning to detect all these errors. Then the next application is market research insights, that is, to know what customers want, how they behave, where the market is going, market research. It's an application like Insights. It is an application of Unsurpassed Learning. This is called Market Research Insights, which is one of the applications that helps us make smarter business decisions. Next up is our reinforcement learning. Okay, reinforcement learning is actually a trial and error method. Okay, that is in a dynamic environment. For example, we commonly use it for gaming. For game claims. Okay, so when we are playing a game, we think that we have crossed a level, and when we cross a level, we get additional coins in the form of rewards. And now, if our car crashes somewhere while we are playing a game, maybe there will be a penalty. So, we get a reward and a penalty or something like that. That is why we use this reinforcement learning. One of its applications is self-driving vehicles and trading algorithms. Okay, now the real world of machine learning. In the real world, where do we use machine learning? Email protection, as we mentioned earlier, when an email arrives in our inbox, we classify it as spam or not spam. What do we use for that? We use a technique called machine learning. Okay, when the email that comes in is correct, our email system tells us whether it is spam or not spam. It tells us 100% accurately. If it is spam, it goes to the spam box. We can't see it. It goes to spam. That's what we call machine learning. One of the great applications is personal recommendations. Personal recommendations is an example of that. Spotify is an app that we use to listen to music. There, too, we get personalized recommendations. Spotify suggests music that is similar to the music we watch or listen to. Similarly, when we use Amazon, it suggests similar products to the product we purchased. For such personal recommendations, we use machine learning. Okay, Spotify, Amazon, what's going on in all this? What are personal recommendations? Similarly, if we order food using Swiggy or Samo, it suggests similar foods to us. The next time we open it, it suggests it. What is this? This is also an application of machine learning. Okay, then autonomous navigation, self-driving cars, sensors, seconds to navigate, come up with an environment. Autonomous navigation, when we say autonomous, what is automatic navigation? When we say automatic, what is automatic navigation? What does it do? It moves automatically, okay, that is. Like what kind of motion is it, maybe it's just a self-driving car, whether it's a robot or a drone, if it's a self-driving car, what is it? It doesn't move on its own, from one plus to another plus. So all this is moving without human intervention. What do they use? What data do they use? There are sensors, there are maps, and then there are intelligent decision- making systems. This is what actually works there. What is the same thing that comes with this machine learning? Okay, now self-driving cars like Tesla and Waymo are based on this. This is what works. Okay, next is fraud prevention. Okay, if we say fraud prevention, now it's the same example we said. We are sitting in some other country, but we have used our debit card and an amount has been debited somewhere else. That's something unusual. Pattern is what we can do there, what can we do there, what is this and what is one of the applications of it. Okay, now there are a few steps to that process called machine learning. Okay, we said that we give data and we give answers along with it. We develop a model. This is actually a method of how it works. But there are many steps to make it work. What I am going to tell you here is a step-by-step road map. The first one is to define our problem. What is our problem? If we are taking a data now, what kind of data is that data? What are we going to predict using that data? We must first understand correctly. Okay, and the next thing to do is to prepare your data. So once we have defined our problem, we should prepare our data. There are many, many things to do in that time. So the first step of data preparation is data collection. Okay, data collection. We have been able to collect data in many ways. When we say many ways, there are many websites where we can collect data. We can download the data set from many websites or many machine learning repositories and start doing our work. Apart from that, we can download the data through web scraping. Okay, there are many techniques like that. So, since this collected data comes from many sources, each data will have a different structure. Okay, some data will not be filled properly. Some data is not visible. Some roses may not have data. Some may not be clean. Some of it may be unwanted data, so there may be many things in that data. Maybe it contains information we don't need. Okay, so we're probably getting unorganized data. So, what should we do first with the data we have received? In that process of cleaning, the first thing we need to check is whether there are any missing values ​​or any missing values? How to handle missing values. So we have many techniques to handle it. Okay, so we replace the missing values ​​or handle the missing values ​​by looking at what or what type of values ​​are missing. Okay, there are a lot of varieties in it. There are many types of data. We handle each data type according to its type. Okay, so we need to see if there are missing values. We need to handle them. Then we need to see if there are any duplicate values. Okay, and if there are any, what should we do to handle them? Then what is called outliers? What is called an outlier? Actually, it is a problem. Okay, we need to remove outliers. If we say outliers, it means nothing. We call outliers something that is very far away from our data. Now, what is the score obtained by the children in a class? Now, let's imagine that a child in that category has a score of 40. All the children in that class have got a score within 40 to 60. Let's say that only one child in that class has got a score of 99. That child who has got a score of 99 is called an outlier. Okay, that is a point that is far from our range of data. An outlier is something that we call odd one out. So if there are such outliers, we need to remove them. So after all this, let's recheck how much the quality of our data has improved. Okay, so this is the step that says prepare data. Next comes the step of selecting algorithms. Okay, then when we say select algorithms, there are many algorithms in machine learning. Okay, there is an algorithm called support vector machine. There are many types of algorithms. So our What is the problem? We will choose algorithms that are suitable for that problem. Okay. What do we do with those algorithms later? We develop a machine learning model. Okay. Then we can understand how much performance the developed model has. Okay. Again, I will connect this as an example. When we study in school, our teachers give us notes. After we study it, after a month or two, what will happen? An exam will come. Why do we give exams? How much have we learned? How well are we performing? Then when the exam comes, we write down all the things we have learned. After writing, the result comes. In the result, we can understand how well we performed. Similarly, here, after the machine develops a model, we can calculate how much performance or accuracy of that particular model. Okay. So, if we calculate the accuracy and the score we get there is terrible, okay, if we say that the mark we get is terrible, I can tune that particular model. Okay, tune it and there are a lot of parameters in it. Then I can adjust the parameters and try to increase the performance. Okay, again, if we connect it to the example we mentioned, and that child who has taken that test and has taken the test will think that the score is terrible. Think of it as a safe score of 54. So, the teacher is thinking, okay, let's give this child a chance. Maybe we can give him a retest. What do you think? The teacher is thinking, okay, let's give that child a chance. Let's tune that child. We provide various types of training to that child. Okay, maybe a teacher tells that child to take a seminar in class. It's a kind of It is a teaching technique or give the child additional flowers and teach him or give him a lot of question papers and teach him what to say. Then we use various techniques and tune it. After that, we retest and its accuracy becomes a little better. Okay, what do we do here? We will calculate the performance of the model, that is, the accuracy of that score. If the score is low, we will tune it and try to make the score better. Okay, then through this tuning, we can improve the performance of that particular model . Once the performance is improved, what do we do next? Deploy the solutions where we need them, where we need them in the real world? We have already defined a problem. So, using that problem, we can implement our product wherever we want. Okay, there are mainly five steps. First one is to define what our problem is. Second one is to prepare the data and select the algorithms. Improve the performance. And then deploy our solutions. Okay, so this is a five-step roadmap that we need to follow while doing machine learning. Now, if we are talking about the case of beginners, we have many no- code platforms available. Among them, Weka is a no-code platform. When we say that Weka is a no-code platform, what do you mean? Its full form is Environment for Knowledge Analysis. It is a no-code platform, meaning there is no code and only very small amounts of code are used. It is a machine learning platform. Okay, so we mainly use this for data preprocessing, i.e. data cleaning and data preparation, and then in machine learning, there are a few categories that come in supervised learning, classification regression, and clustering is a technique that comes in unsupervised. To do all that, we can use a platform called Vaka. So, we don't need to write any program here, there is already a GUI for it. We can upload data sets here. In what format? When we say data set, it comes in many formats, CSV, Excel, and ARFF. There is a format called. It will come in a format like that. So let 's upload the data sets here, okay, and we can run machine learning algorithms with just a click of a button, and we can see the results. Okay, let's just start using no-code platforms like this. Okay, let's get a complete idea of ​​how it all works. Okay, then the next thing we need is a Python program. Okay, in Python programming, we have a library called Cyclic Learn, which is a very, very important library for machine learning. There are many algorithms inside Cyclic Learn. When we say algorithms, if we just put a function, this is the code. Okay, what we need to import here is some steps that come in the data preprocessing, there are import and model building, etc., using the Cyclic functions in it, we can create a very simple Nuke. Okay, but we need to understand why and how they use the particular algorithm. Then we can use learning resources. Okay, Microsoft has a curriculum for beginners available on the platform itself, and in the case of Google, there are many interactive tutorials available. After using all that, we can start at the beginner level. Okay, then how do we overcome this overkill learning? So first of all, we start by setting practical projects. Okay, if we just go with theory, we cannot learn it completely. Saying theory is very important, but if we leave practicals after studying theory, nothing will happen. While studying theory, we should also study practicals. When we say theory, each algorithm has its own math behind it, probability, and every math behind it. There is everything, we should know it completely, okay, okay, when we are doing it practically, we can understand it completely and then use the interactive tools and then, as we said, we can just start using this app. Okay, learning is a hydrating process. We can learn it completely one by one. Okay, build a portfolio. What is a portfolio? Manage our resume properly. Manage your profile. Manage your GitHub account. Then, all the projects you do are completely added to your GitHub account. Okay, if there are projects, then projects are very mandatory. Then, you can add them to GitHub by doing these projects. In adding links, connect with a lot of professionals and connect with professionals who match your profession. Every project you post will be seen by those professionals. Then the calls coming to you or the contacts coming to you will increase. Okay, dad, what can we do through this? We can get into the particular job that we are aspiring for. Dad, let's build our portfolio in that way. Okay, why is machine learning today called machine learning? The demand is very high. Okay, what are machine learning, data science, data analytics, etc.? They are all the same area. Okay, dad, it is not the same area. But there are some relations. Okay, you work completely with data. When you say machine learning today, high- earning careers come in this. Okay, if you say it at the entry level, then you get a lot of slack at the entry level. Salary and S varies depending on the company we join and its size, but still the income we get there will be very high and your career will improve step by step. So while you improve, you can get a good career growth. Machine learning is an area where you can get a lot of projects that provide 100% of your time. Okay, you can do a lot of projects. Okay, you don't have to copy and paste. There are many tools available for you. There are many AI tools available. You can never copy and paste. You can use it. But never blindly copy and paste. Okay, you can do all the things. Okay, okay, so one of the best high-earning careers comes in this area. Okay, and there is nothing that is very essential for our future economy. Machine learning is what we call the world right now. The economy is moving towards what we call the world. And businesses are now using machine learning to make better decisions. Okay, governments are also investing heavily in AI. Whether it's in the case of defense, education, energy, healthcare, etc., what's coming? Our government is investing a lot now. Okay, it's so important. In many areas, our machine learning has created a revolution. If it's in the case of healthcare, as we said earlier, it can predict medical diagnoses and provide personalized medicine. Robotic surgery is used to detect fraud in the case of finance. It can predict stocks. Algorithmic trading is used for all that. Okay, then Personalized recommendations are all about machine learning. Machine learning and all that. As I said, it is a branch of AI. Okay, so knowing all this will be mandatory in the coming days. Okay, so that's the importance of machine learning. Okay, so if you have any queries about our webinar, you can ask now. Okay, Hana, Hi and Nabeel, there are two comments asking for a video related to data analysis. I asked you to do a video related to data analysis. So, let's do a webinar related to data analysis right away. Okay, okay, if you have any other queries, you can post them in the chat. Okay, okay, so thank you.