1 of 13

Predicting Vehicle Prices in Ethiopia: A Data-Driven Approach

Leveraging Telegram Data and Machine Learning for Price Transparency

​

2 of 13

GROUP MEMBERS

Hermela Lemma GSE/4034/16

Kalkidan Kassahun GSE/9049/16

Mignote Deresse GSE/3444/16

Tensae Befikadu GSE/7621/16

Bereket Tamirat GSR/0935/16

​

3 of 13

Introduction

Overview of the Ethiopian Vehicle Market

  • Ethiopia's growing economy is reflected in its expanding automotive sector.
  • Increased demand for vehicles due to urbanization and a rising middle class.

Importance of Understanding Vehicle Pricing

  • Vehicle ownership is often seen as a status symbol and a necessity for mobility.
  • Price stability is crucial for consumer confidence and market health.

Purpose of the Project

  • To explore the challenges affecting vehicle pricing in Ethiopia and suggest
  • To highlight the implications of these challenges for consumers and the automotive industry.

​

4 of 13

Project Overview

  • Objective: Predict vehicle prices using historical data and machine learning.
  • Key Steps:
    • Data Collection
    • Data Cleaning
    • Feature Engineering
    • Model Training & Evaluation
    • Deployment

​

5 of 13

Challenges on Vehicle Prices in Ethiopia

  • Highly Unpredictable Market: Vehicle prices in Ethiopia fluctuate significantly and lack a clear, consistent pricing standard.
  • Lack of Centralized Information: There isn't a reliable platform to check current vehicle prices.
  • Seller-Driven Pricing: Sellers often set prices arbitrarily, leading to inconsistent and potentially unfair transactions.
  • Limited Government Regulation: The market lacks sufficient oversight or intervention to stabilize pricing.
  • Consequences: Buyers face uncertainty and potential overpayment, while sellers lack price guidance.

6 of 13

Towards Price Transparency: Our Prediction Model

  • Goal: To develop a model that can predict vehicle prices accurately and provide a fair price range.
  • Approach: We use a combination of data scraping, cleaning, and machine learning techniques.
  • Benefits: The model provides a reference price range, helping both buyers and sellers make informed decisions.

7 of 13

Gathering Vehicle Data: From Telegram to Spreadsheet

  • Telegram Channels: We identified Telegram Channels where vehicle sales advertisements are posted.
  • Telegram API: We utilized the Telegram API to programmatically collect the data.
  • Data Extraction: We extracted key information from the advertisements, such as vehicle model, year, mileage, and price.
  • Why Telegram?

8 of 13

Preparing the Data: Cleaning and Preprocessing

  • Data Cleaning: Removing irrelevant information and handling missing data.
  • Handling Blank Records: Strategies for dealing with incomplete entries.
  • Outlier Management: Identifying and addressing extreme or unusual data points to avoid skewing our model.

9 of 13

Building the Prediction Model: Using Regression

  • Regression Models: We used regression algorithms to learn the relationship between vehicle features and price.
  • Specific Algorithm: We utilized the Random Forest Regressor for its robustness and ability to handle complex relationships in the data.
  • Model Training: We trained our model on the cleaned and preprocessed data using the Random Forest Regressor.
  • Random Forest Strength: Briefly explain that Random Forest Regressor is a powerful model suitable for these situations.

10 of 13

Model Performance: How Well Did We Do?

  • Mean Absolute Error (MAE): 321737.76
    • Interpretation: Explain that on average, the model's predictions are about 321,737.76 ETB birr off the actual prices.
  • R² Score: 0.938
    • Interpretation: A value close to 1, and 0.938 indicates that the model captures 93.8% of the variability in vehicle prices. A high score suggest a strong model

11 of 13

Deployment and User Interface

  • Streamlit app for user-friendly interaction.
  • Dropdowns for selecting car features.
  • Predicted price range displayed dynamically.

12 of 13

Future Work & Conclusion

Looking Ahead: Future Improvements

  • Data Updates: Regularly update the data to maintain accuracy.
  • Model Refinement: Continue refining the model to reduce prediction errors.
  • Additional Features: Consider adding more features (e.g., vehicle condition, location, etc.).
  • Potential Application: Explore the possibility of creating a user-friendly application for wider accessibility.

13 of 13

Thank You

Any Questions??