Crop-Product Export Value Prediction

Overview

This project aims to forecast the export value of crop products for a specific geographical region up to three years into the future. The prediction model is built using a Multilayer Perceptron (MLP) neural network. The following steps are carried out to preprocess the data, engineer features, and train the model:

Key Features:

• Data Preprocessing: Cleaning, normalization, and handling missing values.
• Data Integration: Merging 13 datasets related to food and agriculture into a single comprehensive dataframe.
• Feature Engineering: Creating lag features, normalizing data, and encoding categorical variables to enhance model performance.
• Modeling: Building an MLP model to predict the export value of crop products three years into the future.

Data Preprocessing

The raw data consisted of multiple CSV files with various categories of variables related to food and agriculture. Data preprocessing was carried out as follows:

• Data Loading: Each CSV file was loaded into a pandas dataframe for examination and further processing.
• Cleaning: Unnecessary columns were dropped, and missing values were handled appropriately.
• Merging: The datasets were merged on the common columns of Country and Year, ensuring that all relevant features were aligned.

Feature Engineering

Several important features were engineered to improve the model's predictive power:

• Lag Features: Lag features for export value were created, including export values from 1, 2, and 3 years prior.
• Normalization: A RobustScaler was applied to normalize the features, improving the stability of the model.
• Encoding: Categorical variables, such as country and product type, were label encoded to allow the model to process them effectively.

DataFrame Structure

The final dataframe used for modeling contains the following features:

• Year: The year the data was recorded.
• Country (Area): Encoded representation of the country.
• Products: Encoded product types.
• Exchange Rate: Average exchange rate for the given year.
• Land Temperature: Average annual land temperature in the region.
• Export Value Lag1: Export value from one year prior.
• Export Value Lag2: Export value from two years prior.
• Export Value Lag3: Export value from three years prior.

Model

The project uses a Multilayer Perceptron (MLP) to predict the export value of crop products. The model is trained on historical data and attempts to forecast the export value up to three years in advance.

Model Parameters:

• Input Features: All the features described above, including the lag features and external factors like exchange rate and temperature.
• Target Variable: Export value of crop products.

How to Use

Prerequisites:

• Python 3.x
• Libraries: pandas, numpy, scikit-learn, tensorflow (or keras), matplotlib

Running the Notebook

1. Clone the repository:

   git clone https://github.com/your-username/crop-product-prediction.git