How CarDekho uses Machine Learning

India has a large list of buyers and sellers of the used cars. Today cars have become our need so most of the people today a have a dream to buy a car. But due to lack of some amount people end up buying used cars. On the other hand, people who need money are selling their cars. But the major question is that what should be the price of a car? Machine learning algorithms have built the solution for this problem, by using attributes of a car we can now predict approximate price of a used car. However, the actual price may depend an algorithm to algorithm but we can somehow manage to reach to the approximate price.

CarDekho is review sharing gateway which shares images, videos and reviews of different cars which are available in the Indian market for sale. This Gateway also has a section of used car which we will consider for today’s case study.

Everyday many used cars are being sold and we are having data of all these cars which contains attributes of the cars such as fuel efficiency, performance, drivability, handling, ride comfort, cO2 emission, sound system, cost, km driven, yeah of purchase and many other attributes which are enough for machine learning algorithm to predict the future price of a used car.

After having the data set ready now, we are all set to apply machine learning algorithms and predict the price.

Car price prediction problem can be solved by using a type of machine learning known as supervised learning in which regression take place. By feeding the past data, who is input contains all attributes of the cars listed above and output contains the price we can solve this problem very easily.

First of all, let’s talk about Random Forest Regression which is a supervised learning algorithm that uses and symbol learning methods for regression. Random forest is a bunch of several decision trees. It estimates the maximum number of decision trees to fit the data set and hence improving the predictive accuracy of the model. It is a good algorithm which can be used to predict the price of used car.

But there is another algorithm which outperforms random forest for predicting the price of a car. We will talk about that algorithm at the last, but let’s talk about another regression algorithm known support vector regressor also known as SVR.

Support Vector Regression is a supervised machine learning algorithm which predicts the discrete values. It finds the best fit line in the hyperplane that has maximum number of points satisfying the condition.

Now let’s talk about the algorithm which is mostly used in car price prediction. It is none other than K Nearest Neighbor. KNN regression algorithm approximates the association between independent variables and uses features to predict the values. Although we have many good algorithms but it is quite obvious that KNN performs good in predicting prices.

And not just predicting prices CarDekho also uses machine learning algorithms in the chatbots. It may use Multinomial Naïve Bayes algorithm and Natural Language Processing which are very important components to build any AI chatbot.