Tutorial 36- Logistic Regression Mutliclass Classification(OneVsRest)- Part 3| Data Science

Key Takeaways

hello on my name is krishna and welcome to my youtube channel so guys today we will be discussing about the logistic regression part 3 and here we are actually going to discuss how we can solve a multi-class classification problem with the help of logistic regression now probably you may have heard only solving binary problem state and binary classification problem statement with the help of logistic regression but now today in this particular video we will try to understand one concept which is called as one versus rest or it is also called as one versus all how does it actually work now if I am talking about one versus rest in logistic regression we know that we actually try to solve a binary classification problem now suppose if I have a problem statement over here and suppose if I have two categories which looks like this right so this is my two classes now in order to solve this particular binary classification problem I try to draw a best fit line in between them and then we try to solve this particular problem right now what is if my output categories are multiple categories like this suppose I have over here a star suppose this is one of the things this is my another classification category right and suppose this is my another category now suppose if I have this kind of solved problem statement can we solve with the help of logistic regression and yes it is definitely possible there is something called as one versus rest now in one versus less what we do is that what what the this particular model will do that is that it will try to you know group this two like suppose there are three categories over here at one time it will try to group these two categories as one category like suppose these two categories will be treated as negative category this category will be treated as positive category now considering this the model will actually create m1 model okay now in the next catechin the next iteration what will happen is that in the next thing is that another model m2 is actually created and now what it does is that in the m2 category first of all it will consider this two categories as one okay and this separate category as a positive category okay in the second case now based on the number of categories again since there are total three categories in the next iteration modern m3 will be created where in these two categories will be treated as one and this particular type category will be treated as a separate positive category so like this combination will be taken up now if I consider this particular example you can see over here suppose I have feature f1 f2 f3 this is my independent feature I have an output feature away as Oh 1 o 2 O 3 so these are all my output features based on the input i1 i2 i3 the output is got as oh one okay similarly if my input is I four i-5 i6 my output is over 2 so like this all the records I have actually having away now how this all categories will get splitted now what will what it will do is that now I have suppose three categories o 1 o 2 O 3 okay now considering this scenario what I told I told you that whenever I have oh one that will be treated as plus 1 okay then I have o2 and o3 will be treated and minus 1 similarly for o 2 in this particular case this will be minus 1 this will get treated as plus 1 and again - 1 o 3 minus 1 minus 1 this will be treated as plus 1 so like this all the combination will be taken up now here you have plus 1 then again here you have minus 1 minus 1 now similarly in o2 you have over here as minus 1 plus 1 and minus 1 now considering this logistic regression when we are using 1 versus rest the first model that it will create it they are going to take this independent features all the independent features right apart from that they will take this first output one feature which will be my output feature and based on this based on this they will create the model m1 ok so the model M 1 will be able to find out whether it is o 1 whether 1 is positive or it is negative or not okay now similarly in the next model that is what it is going to do is that it is going to take this input feature and the output feature will be taken in this particular column that is my ode and for this again the next model M 2 will get created okay and similarly for this particular data finally my output this will be my m3 model now you may be considering that if I give a new test data how will the model actually find out how the urge regression find out the output so whenever I'm giving a new test data the this whole model this m1 m2 m3 model first of all this whole input feature will go to the m1 model and when it goes to the n1 model it will give us one probability value suppose this gives us 0.2 0.1 gave me the answer is point two zero which is my probability that my m2 model basically will give another probability like point three zero okay I can say it as point two five suppose okay Oh point two five then the same data set will be given to the m3 model and suppose the output is somewhere over here is point five five now you can see the sum of all this thing is actually one okay then it will go and check which is the given the highest probability suppose my m3 model is giving the highest probability of 0.5 that basically means my new test data new test data belongs to all three category because my final output that is this oh three is basically having the highest probability okay from all this to that we actually company so this is how one versus rest is actually implemented independently will be creating m1 m2 m3 model and we are not doing it internally the logistic regression is working on that and for that to set this particular parameter there is some parameter called as multi class and there you have to set it as OVR okay when you are doing the coding mechanism it is pretty much simple i will also give you a link where you can see that how it is actually implemented there is just one parameter called as multi class we have to set it as oviya automatically to treat you can solve any classification problem in that particular way and remember for every new test data how many number of categories we have that many number of output values will be getting which will be having the highest probability you can consider that as the output so for my new test data I have got this kind of output you can see that 0.55 is the highest that basically means that it is belonging to the oh three category so this is how the logistic regression one versus rest is actually implemented I have already uploaded all the videos regarding logistic relation this was the part three so I hope you like this particular video please do subscribe the channel if you're not already subscribe I'll see on the next video have a great day anyone lullaby