Multi-Layer Peceptron for Predicting Export Crop Values

Overview

This project involves the development of a machine learning model aimed at predicting the export value of specific groups of crops three years into the future. The model is built using a Multi-Layer Perceptron (MLP) regression model, which was trained on historical crop data. The objective is to forecast future crop export values for different geographic regions, enabling better planning and decision-making.

Table of Contents

1. Performance Evaluation
2. Model Architecture
3. Features and Labels
4. Data Preprocessing
5. Usage Instructions
6. File Structure
7. Contributing
8. License

Performance Evaluation

The model's performance was evaluated using Mean Absolute Error (MAE) and R² score. The MAE was approximately $310,600, which is a small percentage of the target variable given the scale of export values. The model achieved an R² score of 0.864, indicating a good fit, though there is evidence of overfitting as the training score was 0.968.

Visual Analysis

• Scatter Plot: Shows the relationship between actual and predicted values, with points clustered around the ideal prediction line.
• Density Plot: Illustrates that the model’s predictions closely follow the actual data distribution.
• Residual Plot: Displays residuals mostly centered around zero, indicating accurate predictions with minimal systematic bias.

Model Architecture

The model was developed using Scikit-learn’s MLPRegressor, consisting of:

• Input Layer: Receives the features of the dataset.
• Two Hidden Layers: Each with 100 neurons, using the Tanh activation function.
• Output Layer: Produces the predicted export value using an identity function.

Overfitting Prevention Techniques

• Cross Validation: Implemented with a Time Series Split to preserve data order.
• L2 Regularization: Used to prevent weight values from becoming too large.
• Early Stopping: Training was halted if performance did not improve after 10 iterations.
• Bagging Regressor: Employed to improve robustness and reduce overfitting by averaging predictions from multiple models.

Features and Labels

Labels

The target variable is the 'Total Export Crop Value in 1000 US$_lead_3', representing the export value shifted three years ahead.

Features

• Total Import Crop Value
• Fertiliser Use x Pesticides Use (Interaction Term)
• Total Export Crop Value (Lagged by 3 years)
• Area (Geographic Region)
• Year
• Item (Specific Crop Type)

Additional features initially considered were dropped due to multicollinearity.

Data Preprocessing

Steps Involved

1. Data Preparation: Loading datasets, removing duplicates, and handling missing values.
2. Mapping and Merging: Integrating multiple datasets based on common attributes like Year, Area, and Item.
3. Exploratory Data Analysis (EDA): Analyzing distributions and applying log transformations to normalize skewed data.
4. Encoding Categorical Features: Using One Hot Encoding to convert categorical variables into a numerical format.
5. Numeric Feature Transformation: Applying log transformations and scaling to ensure equal contribution of features to the model.

Usage Instructions

1. Clone the repository to your local machine:

   git clone https://github.com/yourusername/crop-export-prediction.git

2. Install the required dependencies:

   pip install -r requirements.txt

3. Run the notebook file.

File Structure

• train_model.py: Script to train the MLP model.
• evaluate_model.py: Script to evaluate the model performance.
• data/: Directory containing the datasets.
• models/: Directory to save trained models.
• notebooks/: Jupyter notebooks for exploratory data analysis and feature engineering.

Contributing

Contributions are welcome! Please fork the repository and submit a pull request for review.

License

This project is licensed under the MIT License. See the LICENSE file for details.