Crop Failure Detection using Bayesian Networks

This project predicts the risk of crop failure using a Bayesian Network model built from agricultural and environmental data. It models how factors like rainfall, temperature, soil quality, pest infestation, and fertilizer use interact to influence crop yield outcomes.

Live Demo: Crop Yield Oracle AI

The live demo is a web-based interface where users can input environmental conditions and receive predicted crop yield probabilities. It serves as a practical demonstration of the backend model, making the research accessible to non-technical stakeholders such as farmers, policymakers, and agricultural planners.

---

Project Overview

Agricultural productivity plays a central role in food security, economic health, and sustainable land use. Traditional models often struggle to capture the complex, probabilistic interactions among the many factors influencing crop outcomes.

This project applies Bayesian Networks to explicitly model the causal and conditional relationships among key variables affecting crop performance. Using this approach, we can:

• Estimate the likelihood of different crop yield levels (Low, Medium, High) under varying environmental conditions.
• Identify the most impactful interventions, such as fertilizer application or pest management.
• Provide interpretable, data-driven recommendations for agricultural decision-making.

---

Dataset

The project uses a synthetic dataset simulating agricultural conditions. It includes:

Variable	Description	Type	Values
Rainfall	Rainfall amount during the crop cycle	Categorical	Low, Adequate, Excessive
Soil_Quality	Fertility and structure of the soil	Categorical	Poor, Average, Good
Pest_Infestation	Whether pests were observed	Binary	True, False
Fertilizer_Use	Whether fertilizer was applied	Binary	True, False
Temperature	Temperature during the growing period	Categorical	Cool, Mild, Hot
Crop_Yield	Final outcome of the harvest	Categorical	Low, Medium, High

---

Key Features

• Bayesian Network modeling using pgmpy
• Conditional Probability Table (CPT) calculations
• Scenario-based probabilistic inference
• Network structure visualization with NetworkX
• Heatmaps to explore variable dependencies
• Scenario comparison charts showing predicted outcomes under different conditions

---

Technical Stack

• Python
• pgmpy for Bayesian Networks
• pandas for data manipulation
• matplotlib and networkx for plotting and visualization

---

How to Run

1. Install required packages

   pip install pgmpy pandas matplotlib networkx
   

2. Load the dataset

    import pandas as pd
    data = pd.read_csv('crop_data.csv')
   

3. Build the Bayesian Network

   from pgmpy.models import BayesianNetwork
   from pgmpy.estimators import MaximumLikelihoodEstimator
   
   model = BayesianNetwork([
   ('Rainfall', 'Soil_Quality'),
   ('Rainfall', 'Crop_Yield'),
   ('Soil_Quality', 'Crop_Yield'),
   ('Temperature', 'Pest_Infestation'),
   ('Temperature', 'Soil_Quality'),
   ('Pest_Infestation', 'Crop_Yield'),
   ('Fertilizer_Use', 'Crop_Yield'),
   ('Temperature', 'Crop_Yield')
   ])
   
   model.fit(data, estimator=MaximumLikelihoodEstimator)
   

4. Perform inference

   from pgmpy.inference import VariableElimination
   
   inference = VariableElimination(model)
   result = inference.query(variables=['Crop_Yield'], evidence={'Rainfall': 'Low'})
   print(result)
   

5. Visualize the network

   import networkx as nx
   import matplotlib.pyplot as plt
   
   G = nx.DiGraph()
   G.add_edges_from(model.edges())
   
   plt.figure(figsize=(10, 6))
   pos = nx.spring_layout(G)
   nx.draw(G, pos, with_labels=True, node_color='skyblue', node_size=2000, edge_color='gray', font_size=12)
   plt.title("Bayesian Network Structure")
   plt.show()