Used Car Price Prediction & Market Analysis

Project Overview

In this project, we assumed the role of Data Analysts and Data Scientists to develop a strategic solution for optimizing used car sales. Using real-world data from second-hand car listings in Spain, we built a predictive pricing model with Azure ML Automated to estimate the optimal market price of a vehicle based on key attributes such as:

• Brand
• Model
• Mileage
• Horsepower
• Year of manufacture
• Location

Additionally, we conducted data extraction, preprocessing, exploratory analysis, and visualization to provide valuable insights into market trends and pricing strategies.

---

🔑 Key Project Components

1️ ETL & Data Preprocessing

• Cleaned and transformed raw data for accurate analysis.
• Most of the steps in "Model Development Pipeline" where also tested in this process using Python.

2️ Exploratory Data Analysis (EDA)

Identified key insights and pricing patterns using:

• Scatter Plots
• Boxplots
• Histograms
• Hexagonal Diagrams
• Line Charts
• Correlation Maps
• Outlier Analysis & Treatment

3️ Predictive Pricing Model (Azure ML Automated)

Developed a machine learning model to estimate vehicle prices based on historical market data.

🔹 Model Development Pipeline:

• Upload Data Asset in Azure ML Studio.
• Selecting Columns
• Clean Missing Data
• Normalize Data and Feature Transformation
  • OneHotEncoder: Applied to car brands, converting categorical values into binary representation (resulting in 500 columns).
  • LabelEncoder: Used for fuel type and transmission. Assigned unique values (e.g., 1 for manual, 2 for automatic).
  • CharGramCountVectorizer: Split brand names into small character fragments (n-grams) to capture internal patterns.
• Split Data (Train, Test)
• Selecting Regressor
• Evaluate Model

📌 Note: Although we designed and executed the data processing pipeline in Azure ML, all these steps were also developed and tested in Python using libraries such as pandas, scikit-learn, and numpy. This allowed us to validate the results and ensure consistency in preprocessing before integrating it into Azure ML.

Model Performance Metrics:

• R² Score: 0.95212 ➝ This is the key metric, as it shows that the model explains 95.21% of the variability in car prices. Therefore, we can affirm that the model has high predictive power and accurately captures the relationship between the variables and the car price.

• MAE (Mean Absolute Error): 1679.7 ➝ On average, the model makes an error of approximately €1679 in its predictions.

• RMSE (Root Mean Squared Error): 3754.7 ➝ This indicates that the prediction errors have an average magnitude of approximately €3754. Our dataset includes both mid-range cars and luxury vehicles with very high prices. This creates significant variability in the data, which may explain why RMSE is significantly higher than MAE. In particular, high-value cars tend to have larger prediction errors, which impact RMSE more.

4️ Power BI Dashboard

Designed interactive visualizations to support strategic decision-making. 📌 (Dashboard included in Streamlit App)

5️ Streamlit App (Final Report)

Developed an interactive web application using Streamlit, integrating Power BI for visual insights.
Deploy URL: https://secondhandcars-priceprediction-6unveathrc59ywyzqcrhz9.streamlit.app/

---

🚀 Technologies Used

• Python (Pandas, NumPy, Matplotlib, Seaborn, Scikit-learn)
• Azure ML Studio (Automated Machine Learning)
• Power BI (Data Visualization)
• Streamlit (Web App Development)

---

📂 Project Structure

├── data/                 # Raw and processed datasets
├── notebooks/            # Jupyter Notebooks for EDA & modeling
├── README.md             # Project documentation

---

⚡ Optimizing used car sales with data-driven insights!