Linear Regression in Machine Learning Explained in 5 Minutes

Key Takeaways

first off linear regression is an algorithm that every machine learning and - yes must know and it is also a good place to start as you know it initially comes from statistics it is a technique studied to understand the relationship between two variables by setting a linear equation in applied machine learning we borrow a lot of algorithms from many different fields including statistics and use them towards our goal of minimizing the error of our model our making the most accurate prediction possible you can see how statistics are very useful in machine learning since we never have a definitive a hundred percent sure answer it's always uncertain optimizing our confidence by minimizing the error of the model which is where statistics comes in handy if we dive deeper into linear regression we will see that the most basic form has two variables the dependent variable Y and the independent variable X you can see it as the equation of a line which is y equals MX plus B where X is the variable you will like to have the equivalent Y result using a slope M and a y-intercept B found from a line representing your data points in short the slope decides the rotation of the line and the intercept decides its translation so if your data points follow a linear pattern like this you will be able to find those two variables to predict how a dependent Y variable is affected in relation with an independent x variable you can then use linear regression in a way to represent this relation we already know from our examples to forecast new observations of course all your data points won't be exactly on the line but by calculating the distance error between your points and the line you will find the best fit with the lowest total error which will be your final line which is where your machine learning minimization algorithm will come in play finally given the representation is a linear equation making predictions is as simple as solving the equation for a specific set of inputs we're just an x value you will be able to quickly find the correspond ye value by simply solving the equation finally in machine learning we often do regression in a multi-dimensional form but don't be afraid of this it works the exact same way as it does in two dimensions finding a relation between all the variables to predict a future observation as I explained previously linear regression is used to understand the relationship between two variables this way if you'd like to know how an increase in the gross domestic product GDP of a country has an effect on the life satisfaction of its citizen using linear regression could be a great and simple way to get an answer for example let's use this graph taken from a study made in European Union by researchers at Vilnius University using data from 2008 it shows the life satisfaction taken as a country's percentage of respondents saying they are very satisfied or fairly satisfied with the live delete versus the GDP per capita of those countries just by visualizing this graph you may already see that it looks like it follows a line which will look like this from bulgaria to sweden where the increase in GDP per capita clearly increases the life satisfaction of its citizen the line will be calculated by measuring the minimum distance between the points in order to find the most optimal line which will look like this one then if we add a new observation say we have France which had the GDP per capita of around 27,000 euro in 2008 you will have a life satisfaction resulting of around 80 percent which is most likely probable in short linear regression helps you find the relation between multiple variables based on the simple equation of a line [Music]