Bulldozer Price Regression (End-to-End ML Project)

This project is an end-to-end machine learning workflow to predict the sale price of bulldozers using historical auction data.

It is based on the Bluebook for Bulldozers Kaggle competition and follows a complete ML pipeline — from data preprocessing to model evaluation and prediction.

Project Objective

The goal of this project is:

To predict the future sale price of a bulldozer based on its characteristics and past sales data.

Since the output is a continuous value, this is a regression problem.
Additionally, because the data includes time-based features (sale dates), it also involves time series forecasting concepts. :contentReference[oaicite:0]{index=0}

---

Dataset

The dataset comes from the Kaggle competition and consists of:

• Train.csv → Historical data up to 2011
• Valid.csv → Validation data (Jan–Apr 2012)
• Test.csv → Test data (May–Nov 2012, without target) :contentReference[oaicite:1]{index=1}

Key Features

• Machine specifications (ModelID, ProductSize, etc.)
• Usage data (MachineHoursCurrentMeter)
• Sale information (state, auctioneerID)
• Time-based feature (saledate)

Target Variable

• SalePrice

---

Machine Learning Workflow

This project follows a structured ML pipeline:

1. Problem Definition

Predict bulldozer sale price using historical data.

2. Data Exploration

• Load dataset with Pandas
• Understand structure, missing values, and data types
• Identify key features

3. Feature Engineering

• Convert saledate into:
  • Year
  • Month
  • Day
  • Day of week
• Handle missing values:
  • Numerical → filled with median
  • Categorical → converted to numerical codes
• Add missing-value indicator columns

4. Data Preprocessing

• Convert categorical variables into numeric format
• Ensure training and test data have the same feature structure

5. Model Building

• Model used: RandomForestRegressor
• Reason:
  • Works well on structured/tabular data
  • Handles non-linear relationships
  • No need for feature scaling :contentReference[oaicite:2]{index=2}

6. Model Evaluation

The evaluation metric used is:

• RMSLE (Root Mean Squared Log Error)

Why RMSLE?

• Penalizes large percentage errors
• Suitable for price prediction problems with wide value ranges :contentReference[oaicite:3]{index=3}

Other metrics:

• MAE (Mean Absolute Error)
• R² Score

---

Model Optimization

• Used RandomizedSearchCV for hyperparameter tuning
• Tuned parameters such as:
  • n_estimators
  • max_depth
  • min_samples_split
  • max_features

This improves model performance while keeping training time manageable.

---

Predictions

• Preprocessed test data using the same pipeline
• Matched feature columns with training data
• Generated predictions using the trained model
• Created submission file with:
  • SalesID
  • Predicted SalePrice

---

Contact

If you have any suggestions or feedback, feel free to connect!

---