Car Sourcing with Data-Driven Segmentation

• 

  Cluster Profile

• 

  SHAP plot

Inspiration

Carsome is a leading online used car marketplace that streamlines the buying and selling process for individuals through its innovative platform. To optimize their sourcing strategy, it is crucial to identify the most valuable cars to purchase within their budget. Therefore, the goal of this project is to create a robust clustering model that can effectively segment cars based on shared characteristics. By leveraging data-driven insights, Carsome can make informed decisions on which cars to source and list, rather than relying on a generic approach.

What it does

A cluster model was developed to segment cars based on shared characteristics, helping Carsome make informed decisions about sourcing, pricing, and market positioning.

How we built it - Methodology

This project has 5 main phases:

1. Business understanding understanding the business problem that needs to be solved, which was to segment cars based on their features to better understand their value and demand in the market.

2. Data understanding Exploratory Data Analysis

3. Data preparation removing duplicate rows, dealing with missing values. deriving features, selecting relevant features, and scaling numerical data for model input

4. Modeling k-prototype was selected as our cluster model. Elbow Plot and KneeLocator was used in finding the optimal number of clusters

5. Model evaluation Cluster Profilling

Challenges I ran into

One of the biggest challenges I encountered was determining the most effective approach to tackle the problem statement. There were several trials and errors throughout the process before settling on a suitable plan.

Initially, I attempted to segment car dealers instead of cars, grouping the cars based on the dealer's ID and aggregating the data for selected features. However, I realized that the dataset was not suitable for this method as this approach yielded unsatisfactory results. I had to re-evaluate and pivot my approach to focus on segmenting the cars based on shared characteristics, which ultimately led to the development of the current cluster model.

Accomplishments that I'm proud of

I am proud of the perseverance I demonstrated throughout the course of this project. Despite facing numerous challenges and encountering moments of doubt, I never gave up and continued to work towards finding a solution.

What I learned

I have come to appreciate that building machine learning models is an iterative process, and it often involves trial and error. I have learned the importance of experimenting and trying out different approaches until you find the one that works best.

I have enhanced my knowledge and skills in building machine learning models. I learned how to use Azure Machine Learning and perform data analysis in the cloud. I also developed my skills in data cleaning, data visualization, and machine learning techniques.

What's next for Car Sourcing with Data-Driven Segmentation

The successful creation of a data-driven car segmentation model opens up several opportunities for Carsome. Here are some of the next steps:

Deployment of the machine learning model: The developed model can be deployed to automate the process of car segmentation. This can help Carsome save time and resources in sourcing and listing cars on their platform.

Exploring additional features: While the current model uses a variety of features to segment cars, there may be additional features that can further improve the accuracy and effectiveness of the model. These can be explored to create an even more robust segmentation model.

Built With

• azure
• azuremachinelearning
• python