Features and Feature Engineering in Machine Learning - An Introduction

Key Takeaways

hi everyone welcome back to my youtube channel my name is imath and today we will be talking about features and feature engineering this is an introductory video very very basic if you're just getting started in the field of machine learning and you want to know what features are and what feature engineering means this is the video for you in this we will just take a look at how you define features and what the process of feature engineering means in the context of building a machine learning model or a machine learning system so let's get started uh to start off uh i will present a few statements to you that you might have heard while reading online blogs or talking to some of the people who are working actively and trying to build models so statements like what features did you use while trying to build your model or the features you used did not have a lot of protective power or feature or statements like these are very few features you need to add more these are a lot of features you need to reduce them so everywhere you see uh there's a lot of talk of features and people especially while building a machine learning model and then they ask you about what kind of features you used how effective are they how protective are they and and if you if you are wondering what this even means then you've come to the right video to understand what feature mean means and what future engineering means in the context of building a machine learning system so now let's get started with what is a feature so before we look at what a feature is we would i think what's important to understand is the entire process that uh or entire process of building machine learning model is dependent at its core on something we call as data and data at its core is observation of real world around us so whatever you see around you you see people you see machines you see systems you see logs all of these things are in a way what we call data right so data at its core is how you observe the world around you and then store that in a format that could be reused and re-read features is essentially a numeric representation of the raw data that you observe around you so it's a way of representing these observations that could be useful to perform further tasks and that is it right so features are derived in a way from data but they are a way of storing your data i mean you store your data but there are way of perhaps transforming or changing your data that is numeric and that could be useful to perform for the tasks so i have an example for you here and this is this will hopefully make you understand how what what is features uh in the context of building this uh machine learning systems right so you have a real world activity here and perhaps there's a doctor or a nurse uh care of a patient and figuring out different vital information about them and that is stored in a numeric format that could be useful to do a lot of further analysis or prediction that you see on the right so on the left is the process with a real-world activity of a doctor monitoring a patient and on the right you see different columns or different features and when i say features each of these individual columns like the patient gender oxygen level pulse rate all of them could be considered as individual features and based on and each of the row that you see are different patients what's interesting also to note here is that all these values are numeric so the numeric representation of raw data is called features as we saw was what's the definition earlier right so we have a real world activity we store the data and then we have features so that is all about features right features is essentially storing data in a numeric format what is feature engineering so as you might have probably guessed engineering features or or creating features uh is what is called feature engineering and to put it into context i would like to take you through this diagram where you see where feature engineering fits while building a machine learning system right so if you look at this diagram this is a very popular diagram on the internet you have different data sources the source ones are source two so on up to source n and all of these data sources are in different formats with different data types with say different amount of information all of this gets stored as raw data when you're trying to model a certain process and that raw data is then transformed into features and these features are fed into your modeling process and then you get your insights or your predictions so that is the entire end-to-end pipeline and feature engineering sits right in the place between your raw data and the features so essentially the process of converting your raw data into features is called feature engineering and uh you do that in a multitude of ways so here is here it says clean and transform but there are a lot of different things that you do to be able to create those features that are useful for a task that you want to perform so when you're modeling you're essentially trying to perform a certain task you're not modeling without an inherent goal in your mind right so you have a certain task you want to perform and modeling uh is done with respect to that so feature engineering is also in a way dependent on the task at hand it's also dependent on the model you are going to choose and the data you have so these three things are what uh contribute to your future engineering so feature engineering uh is not like an independent activity like some of the people around might think but it is also dependent like i said it's dependent on the data you have it's dependent on the modeling task you have and the models that you want to subsequently use so this is the phase this is the this is an example of how feature engineering is used it's the process of converting raw data to features and now let's take a look at the example i was talking about so earlier we saw this example but i intentionally omitted one part of the process because i was trying to show you features but in the real world you might not directly go from real-world activity to features they might be and and it happens more often than not that you store the real world activity as data uh and then you go through an entire feature engineering process and convert that to features uh like for example in this case right so like i mentioned the the example is about a certain process that the doctor or the nurse is doing and they might they of course they will not go ahead and write patient gender as one or zero because it doesn't make sense to them so they would represent it as m or f male or female in this case but for the purpose of our modeling we need to convert all character and string features into numeric features and i mean all characteristic data into numeric features and that is where the process of feature engineering comes into play so this is one very basic kind of feature engineering where you're basically encoding the different categories that you see in patient gender to a representative numeric value so male is represented as 1 m is represented as 1 and f is represented as 0 in the case of patient gender so that is what you do in future engineering you convert your raw data into numeric values feature engineering uh yeah just to summarize right so basically conversion of raw data into a numeric value is future engineering which engineering is dependent on the data you use is depend on the modeling you want to do the task you have at hand and the models you select eventually for the task so that is what i wanted to talk about give you an introduction to what features are uh talk about the process of feature engineering and where future engineering fits in the entire pipeline and how data can be converted from raw form into a form that is uh useful uh by a machine learning model and also what does feature engineering depend on so all of these things is what i wanted to convey in this video if you like this video please don't forget to like share and subscribe and yeah of course subscribe to my channel because i will be putting out more such videos in the upcoming few weeks and months hopefully until next time see you around