Caroogle
An AI-Driven Automobile Analytics Platform

Caroogle
An AI-Driven Automobile Analytics Platform

Overview

Our client had a brick-and-mortar store, a car dealership business, that buys and sells cars at reasonable prices. It was going pretty smooth until he stumbled upon the hassle of the manual process in collecting car data from numerous sources. It was obviously giving him a tough time in exploring the market, carrying out extensive research, and figuring out accurate automobile prices. He was quite perplexed about the labor-intensive manual process and wanted to automate the whole process.

That’s where we came in. We put forth our teams’ expertise to tackle the core issue in managing the entire business. Our mavens came up with a solid idea of automating the whole car dealership business through an AI-powered analytics platform. Our proposed solution would cater to a bigger and targeted market, enabling our client to generate more profits, and analyze overall growth through intelligent business insights.

Overview

Our client had a brick-and-mortar store, a car dealership business, that buys and sells cars at reasonable prices. It was going pretty smooth until he stumbled upon the hassle of the manual process in collecting car data from numerous sources. It was obviously giving him a tough time in exploring the market, carrying out extensive research, and figuring out accurate automobile prices. He was quite perplexed about the labor-intensive manual process and wanted to automate the whole process.

That’s where we came in. We put forth our teams’ expertise to tackle the core issue in managing the entire business. Our mavens came up with a solid idea of automating the whole car dealership business through an AI-powered analytics platform. Our proposed solution would cater to a bigger and targeted market, enabling our client to generate more profits, and analyze overall growth through intelligent business insights.

Challenge

  • We needed to devise an independent, automated automobile price analytics system, updated by current market prospects.
  • We also needed to figure out the gaps and deficiencies in the existing system, as we wanted to automate them through machine learning techniques.
  • Our proposed model would have to fetch car prices data from multiple websites in order to be accurate, efficient, and flawless.

Challenge

  • We needed to devise an independent, automated automobile price analytics system, updated by current market prospects.
  • We also needed to figure out the gaps and deficiencies in the existing system, as we wanted to automate them through machine learning techniques.
  • Our proposed model would have to fetch car prices data from multiple websites in order to be accurate, efficient, and flawless.

Proposed Solution

  • We designed a robust algorithm that crawls through multiple websites to get automobile data.

  • The scraped data was stored in a centralized database.

  • We applied various preprocessing techniques to clean the data for training our ML model.

  • We used XGBOOST, a tree-based model that classifies the features of cars, carries out an assessment and puts them in three categories: cheap, super cheap, and hyper cheap.

  • We fed the ML model with clean data for training and predicting accurate car prices.

  • In order to balance the level of three categories, as stated above, we had to add extra weights to make the three classes equivalent for our ML model. In this way, model bias towards one class would be eliminated and our model would predict accurate labels for the automobile.

Roadblocks

We needed to fetch real-time and updated data from a wide variety of sources and latency in getting live data and storing it into the database was a problem we could not tolerate. For run-time parallel processing and data fetching, we introduced a Redis server to act as a liaison between the two endpoints. It solved the speed problem like a charm.

The second head-scratcher presented itself in the form of multiple outliers. We noticed the variance in data distribution was not properly skewed. The inconsistency of labeled data among three classifiers cheap, super cheap, and hyper cheap, became a challenge because we needed a balanced and evenly distributed set of data. So, in order to generate unbiased and inaccurate insights about car prices, we came up with our own algorithm to deal with inconsistent records.

One major roadblock was to keep pace with the current changing trends. We had to maintain our pace with a rapidly evolving market that was struck by a global pandemic. External factors play a major role in evolving the model and enable us to change the old statistics and update the model with new data sets. We managed to tackle the segregation of the data sets for pre and post covid times. Model predictions were quite different after we maintained and updated it with new information.

[fusion_builder_container type=”flex” hundred_percent=”yes” hundred_percent_height=”no” hundred_percent_height_scroll=”no” align_content=”stretch” flex_align_items=”