Imputing missing values for Dataset of Used Cars

Abstract

Missing values in a dataset has always been a problem for data analysis and modelling. Building a model over a dataset where the missing values are not handled properly will definitely degrade the accuracy and performance of model. This problem particularly impacts deterministic models. Knowing that majority of the models that are used today are deterministic makes dealing with missing values crucial before applying the machine learning model. In this paper we have discussed various approaches such as statistical method (using mean), MICE and KNN for imputing missing values and tested their accuracy in combination with two prediction algorithms linear regression and random forest regression. We have used dataset of used cars containing missing values in few columns to predict the price of car given the details of car and thus comparing the accuracy of the estimated price with different approaches.