By now you've learned a little bit about what Machine Learning is and the value it can bring to your businesses and started assessing the feasibility of a few ML use cases. In this module, I'll focus on how machine learning works. This will help you understand what you'll need to take into consideration as you manage each of the remaining phases of a machine learning project. I'll start by covering some common ML problem types and the areas we'll focus on in this course. Next, I'll go over the core elements of a specific type of machine learning. These are standard Algorithms, Data characteristics, predictive insights and decisions. Some of these concepts require examples to really be understood. So I'll present a few of those. I'll close the module with why ML is particularly important now and picking up momentum across organizations. There are two main types of machine learning, supervised and unsupervised learning. All the use cases I covered in the previous module were examples of supervised machine learning. What do I mean by supervised? If someone sitting on top of the machine and watching it learn? Not quite. It's called supervised learning because a data scientist or ML practitioner typically acts as a guide to teach the machine what conclusions that you come up with. For supervised learning to work, the machines possible outputs need to already be known and then labeled data is used to train the machine. We will return to this little later in this video. Supervised learning problems fall into one of two categories, regression or classification. A regression problem is when we're trying to predict a continuous numeric value, for example, predicting the price of a house. A classification problem is when we're trying to predict a distinct category or class. For example, predicting whether something is true or false. Unsupervised learning is fundamentally different because it doesn't involve predicting something specific. It's about uncovering patterns or structure in your data. Seeing whether your data naturally falls into different groups or clusters. Let's look at some examples for each of these problem types. Here's what I'll do. I'm going to show you a statement, then we'll determine what type of ML problem it is, regression, classification or cluster and then we'll extract the benefit of solving that problem. This exercise will also help you in the future. As you brainstorm more ways to use ML, it'll help you to identify the ones that will have the most impact. Your first problem statement is to predict how many cars will pass through a specific city intersection between 8.00 and 9.00 A.M. What type of problem is this regression, classification or clustering? If you said regression, you're correct. Remember the clue is in the statement itself. Predict tells us it's a supervised learning problem and how many tells us that we're looking for a numeric value. So it falls into the regression category. What's the benefit of solving for this problem? There are a few benefits. For example, you'd be able to optimize traffic lights according to the number of cars that pass through. Additionally, you can use the same solution to offer users alternative routes on their GPS apps. Let's try another example. Predict the location of each image or video in a repository that contains thousands of images and videos. What type of problem is this? Here's a hint. The word predict should tell you it's supervised learning. So is it regression or classification? If you answered classification, you're correct. The word location is the clue here. It tells you that you're looking for a distinct name or category. Now for the more challenging part, what's the benefit of solving for this problem? One answer is for Automation. On the surface, many problems like this one seems simple. However, the magic happens when you can keep digging for an even higher-value problem statement. This can be achieved when you combine what ML can do with your own domain expertise. Think of the various professions other that might use images or videos. They need to know where images and videos are and also use them to make repeat decisions. How might a solution to this problem enable these professions to do their job? Doctors for example, use images all the time to diagnose illnesses. Investigators use photos to solve crimes or to find missing persons and in terms of improving human life in general, visually impaired people can benefit from classification ML solutions to read handwritten documents or to understand the context in an image or video. Let's look at a third example. I work in a Learning and Development Organization, we built 100s of courses for a variety of audiences. What if we wanted to use the metadata from all of our courses to determine which ones have the most content overlap. What problem type would this be? You guessed it, it's clustering. Solving this problem has two immediate benefits. What might they be? Well, would be able to determine which courses to update and which ones we can deprecate. It can also help us to identify the gaps so we can create new content. Until now, we've looked at very specific problem statements and categorized them as regression, classification, or clustering. Sometimes though, the problem you're focusing on is really big. To solve it, you'd actually need to break it down into smaller problems statements. Let me give you an example we've all become very familiar with. Suppose the question is, how do we stop the spread of COVID-19? How do we cure it? Right away we know that this is a high value problems to solve. Can ML be used to solve our problem? Well, let's break it down by individual problems based on some of what we already know. Can we predict whether a patient is positive for COVID-19 using CT scans or imagery? Can we predict the number of patients who would be infected in a specific region? Can we uncover similarities and differences between those who have already tested positive with the virus? There are many more Sub statements for now, let's go through the list here and identify their types. The first one is classification. Again, the hint here is use of imagery. The next one is regression, because we're looking for the number of patients and the last one is clustering. Now let's look at the purpose or benefit of solving for each of these. For the first one, this is a more efficient way of testing and diagnosing patients. At the time this course is recorded, testing kits takes several hours to yield a diagnosis. The benefit of the second one is to identify the appropriate health care and safety measures ahead of time. You'd want to be able to optimize staffing and room availability based on which regions would be hit the hardest. Thirdly, you want to use the data to make some new discoveries that can inform other decisions. For example, you can identify whether there are multiple strands of the virus. Maybe you'll find trends in the data that paint a clear picture about which groups and regions are most susceptible to the virus. As you learned in this video, Machine Learning is a really big topic. For the purposes of this Fundamentals course, we're going to focus on supervised learning examples when describing the rest of the phases in an ML project.
