Supervised learning explained
Supervised learning is a sort of technique where you say that given a data from the past, that there are many attributes associated with that data set, you also have something called a label. Supervised learning creates a perception of an object which is reinforced by labeling the object helping in identifying not only the object but also its variability in the future.
Learning as a kid
So, for you, as if you were a kid learning to identify different kinds of fruits, for example. You visually look at that fruit and you know what an apple looks like. You form a mental perception around it. And someone taught you that anything that looks like this shape is an apple. Similar is the case with other fruits as well, for example a banana, orange and so on. So this visual perception you have learnt as a kid and the other help you have got from somebody else told you that this visual perception of yours is an apple. This is what is called a supervised learning.
The input to Supervised Learning
There is an input feature to your perception which is more around the color, shape and the structure of that fruit and somebody else telling you that this sort of a thing is something called an apple. So, these two combined, the machine learning model trains itself. Over a period of time, irrespective of the sort of shape and color and textures of different types of apple, you will be able to identify that this is an apple. So, no matter how different tricks you do, no matter how nature plays out in the future as well in coming out with new varieties of apples, your perception is very strong in terms of identifying an apple because somebody has trained you on that. And this is typically what happens in a machine learning model as well.
Training and Accuracy needed
You train yourself with a lot of input data about any given object and based upon that you have a label and this label is what tells you that this is an apple. Remember here that because we are training somebody on what that object is you should be very careful that whenever you curate a data set for a supervised machine learning algorithm your data should be 100% correct. Even if you miss out on 10% of data set where you feel the labeling is wrong, expect that 10% as an error in output as well. Your model is as good as your data in simple terms.
There are many algorithms which build supervised learning. Ensure you learn about them during your Data Science training. For example, if you build a classifier for a fruit, the labels that will be applied – this is banana, this is an apple, this is orange based on demonstrating the examples of classifiers of banana, apple and orange respectively.