Created by Zhou Quan Eric Gao

Problem Overview

The science of training machines to learn and produce models for future predictions is widely used, and not for nothing. Agriculture plays a critical role in the global economy. With the continuing expansion of the human population understanding worldwide crop yield is central to addressing food security challenges and reducing the impacts of climate change.

Crop yield prediction is an important agricultural problem. The Agricultural yield primarily depends on weather conditions (rain, temperature, etc), and pesticides, and accurate information about the history of crop yield is an important thing for making decisions related to agricultural risk management and future predictions.

This repository was created as the basis research for the Tu Bishvat presentation given to find out the best performing algorithm for crop yield prediction; a specific prediction was tested in a specific scenario for an interactive activity, where Country: Australia Item: Wheat Year: 2013 Average Rainfall (mm): 534.0 Pesticides used per tonne: 45177.18 Average Temperature (°C): 17.4 Area Yield (hg/ha): Unknown

Check out specifics in the pdf of the presentation given ML_CropYield_Presentation.pdf

Check out notes taken, papers related to the problem, and a more in-depth overview of the problem from ML_CropYield_Notes.docx

Data Availability

The original dataset can be found here: https://www.kaggle.com/datasets/patelris/crop-yield-prediction-dataset

All datasets (publicly available datasets) here are taken from FAO (Food and Agriculture Organization) and World Data Bank.

http://www.fao.org/home/en/ https://data.worldbank.org/

Models Included

This repository contains various machine-learning models for predicting crop yields based on environmental and agricultural factors.

• Linear Regression
• CNN (Convolutional Neural Network) *Not Included in Model Comparison
• Decision Tree
• Random Forest
• KNN (K-Nearest Neighbors)
• SVM (Support Vector Machine)
• Boosting

Project Structure

Under the Figures Folder:

info.py: produces the graphs Log_Transformed_yield,png and hgha_yield.png Visualization.py: produces the graphs area_countplot.png, item_countplot.png, yield_per_country.png, and yield_per_crop.png

Under the Algorithms Folder:

Boosting.py Gradient Boosting Tree CNN.py Convolutional Neural Network DecisionTree.py Decision Tree KNN.py K Nearest Neighbor LinearReg.py Linear Regression RandomForest.py Random Forest SVM.py Single Vector Machine

Results

Result.py Evaluates the result of all the machine learning algorithm under three metrics

r^2 (Coefficient of determination), r^2.png: the proportion of variance in crop yield that can be explained by the independent variable(s)

Prediciton Specific |Error|, ScenarioSpecific_Pred.png: absolute value difference between the predicted yield by the machine learning algorithm and the actual yield

Theoretical, Theoretical.png: Theoretical limits the machine learning algorithms can reach with optimal tuning and allowed maximal run time

Overall, based on the three metrics, the ranking for the algorithms for this specific study is as follows:

1. K Nearest Neighbor
2. Random Forest
3. Decision Tree
4. Gradient Boosting Tree
5. Linear Regression
6. Single Vector Machine