Price estimation of pre-owned automotive vehicles

Overview

Client is a pre-owned automotive car sales showroom who were keen on modernizing their business process. They were looking for an automated price estimation system that could generate car prices depending on the market condition. They were looking to build a website where new customers can login to their website, register with their pre-owned car details and the system would then generate and produce an automatic price estimate.

Challenge

We were tasked with the goal of developing a price estimation model for any pre-owned car that a new customer might want to sell on their website. The challenge here was to analyze large data pertaining to various parameters about the car, namely, make, model, year of manufacture, mileage, its previous owner profile like gender, age group. We also had to match this data to the current market conditions and develop an automated system for price estimation. The automatic price estimation process had to be a touch-free process for the customer and also free up the manpower requirements on the car sales process.

Solution

To develop the price estimation ML model, we trained it with a dataset of past sales records. We collected the data about the car details of the previous few years of car sales, the sale price of the car along with the buyer and seller customer profiles. We converted the car sale details, make of the car and model details, the car buyer details, and market condition details into a multi-feature dataset to make it amenable to ML training. The price estimation problem is a type of Regression problem in Machine Learning. We studied the performance of various Regression algorithms like Linear, Ridge, SVR, and MLP Regressor. We observed that the MLP regression performed the best in terms of both accuracy and execution time. We packaged the ML model as a cloud function which ingests the new car details and generates a price estimate of the car. The cloud function exposed an HTTP API that was integrated into their new e-commerce website.

Impact

We were able to help our clients eliminate the manual effort involved in the car sale price estimation process and also increased the conversion on their website by 35%. This project was developed within a 6-week time frame. The ML Regressor-based ML model was predicting pre-owned car prices at an accuracy of 95%. The client was able to turn out quicker sale cycles and increased revenue

Technology stack