White Box Vs Black Box Models In Machine Learning- Data Science Interview Question

Key Takeaways

hello all my name is krishnak and welcome to my youtube channel so guys today in this particular video we are going to understand the differences between a black box model and a white box model i hope you have heard this kind of terminologies and remember guys this particular question was recently asked in an interview to one of my subscriber and again he got confused while explaining this particular topics so he requested me to make this specific video that is the reason why i'm making this video guys and before we proceed please do hit the subscribe button press the bell notification icon because we are about to reach 400k subscribers uh this will definitely motivate me if you are subscribing um my channel and definitely i'll be uploading many videos as we go ahead right so let's proceed and let's understand the basic differences between a white box model and a black box model so i will ask you one specific question okay if i consider a neural network okay just answer this question that whether it belongs to a black box model or a white box model suppose if i'm using a neural network and creating a specific model let it be a classification model a regression model which category does it belong to if you know the answer pause the video write the comment in the comment section write the answer and once you complete this particular video just try to check whether this particular answer is correct or not right so let's proceed now let's consider a black box model what is this black box model let us consider neural networks okay and you know that neural networks can be used for different different problem statement for regression classification image classification object detection anything right now when we are using this kind of neural networks we have many many layers we have many many neurons internally so many weights are there and when we train this kind of neural networks it is very very difficult to visualize how these weights are actually changing right because obviously because those those neural networks are quite complex we'll also not be able to understand that how the future features are actually interacting within that particular neural network because there are many layers there are many calculations that are going on right and this kind of complex models we actually call it as black box model and now there are some of the properties with respect to this kind of black box model usually this black box model gives us very good accuracy when compared to the other model specifically the white box model i'll also talk about that too the accuracy is quite good the complexity is more in this it will be able to determine the non-linear properties also inside the data set that is the power of this particular black box model you know because if i if i take one more example suppose if i take random forest it is not just only neural network will come in to uh come as a category of black box model if you take some of the um bagging trees or boosting trees like random forage or xg boost internally you select hundred and hundred and trees right and how that trees internally split and divide the features it is very difficult to just visualize so this kind of models are also called black box models some of the properties of black box model is that the accuracy will be very high it will be able to determine the non-linear properties it will be able to solve complex problems the complexity will definitely be high you know when compared to the white box model now let's go ahead and try to understand what exactly is a white box model okay now white box model suppose if you have used some of the algorithms like linear regression logistic regression which are specifically used for some purpose right internally what is happening simple and mathematical equations are actually being used if i talk about linear regression or decision trees also right internally some mathematical equations are actually happening and there you will be able to understand how that particular model is basically created how the features are actually interacting with each other what is the correlation with respect to the feature how that but based on a specific feature how the decision is being made everything you will be able to understand so there is a very big other topic which we basically say it as explainable ai and explainable ai is basically the work of an explainable ai is to understand interpret the model that you have actually created and suppose if you are using the black box model it becomes very difficult to understand that how that specific neural network is actually working yes definitely we will use some kind of performance metrics accuracy this that and all right different different performance metrics will be there but internally if i change this value by this much how will i be able to get the how how the output will actually behave that is very much difficult in black box but in weaker models a white box models this thing you'll be able to see it and that is the reason nowadays there are so many companies who are exploring who are working in this explainable ai part where they are building tools and softwares they are building tools and software or libraries which will help them to basically interpret this kind of models that is what it is doing and it is really really many people are working on there are many startups who are working on it and some of the libraries if i just talk about some of the libraries which have again uploaded in my channel also okay there are libraries like lime shapash right these all libraries actually help you to interpret the models this complex models in an amazing way yeah if you take exubush or random forest they will be able to interpret this in an amazing way i've also i'll make sure that i'll put the video tutorial in the description of this particular video also and now that many startups are focusing because see there are a lot of auto ml tools auto ml library is also being used to do all this kind of machine learning task people are moving towards the automation side at least they are trying to move into the automation side you know and they are trying to build some automated models there also this kind of libraries will be handful so people though companies they are coming up with this amazing libraries which is a part of explainable ai and its main functionality is basically to interpret the models now interpreting the black box model is the most complex part because there are many things that is involved over there and that is what people are actually focusing on if i talk about this lime and chapatis easily the white box model like linear regression logistic regression decision entries they'll be able to interpret it in an amazing way or by by seeing all the diagrams you'll be able to see they'll also give you the information how the features are correlated how the features are not correlated how independent features are correlated with the dependent feature all this kind of information will be given to you and still interpreting a black box model is the task that people are trying to solve because obviously the network the neural network is quite complex we cannot just observe each and every weight of feature how it is behaving and it can be also because of the depth of the neural network how many number of neurons how many number of computations are actually happening right so this was the basic difference between black box and white box model black box model we basically say it as accurate model we this white box model we actually say it as a weaker model okay and obviously the black box model is better than the white box model but there are scenarios if you have less amount of data if you're solving a simpler problem sometimes this white box model may solve better than the black box model they are those kind of scenarios also and that is the reason why i say you when should we use deep learning or when should you use machine learning instead of deep learning that is the question also and i have uploaded those kind of videos so you should definitely have a look on to that so i hope you like this particular video please make sure that you subscribe the channel press the bell notification icon and hit like if you have liked this particular video i'll see you all in the next video have a great day thank you one doll bye bye