Comprehensive Analysis of Crop Yield Prediction and Input Optimization Using Machine Learning

Introduction

This project aims to leverage machine learning techniques to predict crop yield, optimize input usage, and assist in agricultural decision-making. It includes detailed analyses and tools for:

• Predicting the effect of annual rainfall on crop yield.
• Recommending optimal fertilizer and pesticide usage.
• Yield prediction based on state-specific conditions.

A user-friendly interface was developed using Tkinter for real-time interaction and predictions.

---

Features

1. Rainfall Impact on Yield Prediction

   • Inputs: Annual Rainfall, Cultivated Area, Fertilizer, Pesticide, Crop Type.
   • Model: Regression Forest.
   • Metrics: R², MAE, MSE.
   • Insights: Highlights the need for irrigation planning and optimized input usage.

2. Optimal Fertilizer and Pesticide Recommendation

   • Model: Regression Forest with feature engineering.
   • Outputs: Suggested optimal levels for fertilizers and pesticides per crop.

3. State-based Yield Prediction

   • Inputs: State, Area, Annual Rainfall, Fertilizer, Pesticide, Crop, Season.
   • Outputs: Region-specific yield predictions and input optimization.

4. Tkinter GUI

   • Real-time predictions with an easy-to-use interface.
   • Input fields for rainfall, fertilizers, pesticides, and other features.
   • Displays yield predictions and recommendations interactively.

---

Methodology

1. Data Preprocessing

   • Normalized numerical inputs.
   • One-hot encoded categorical variables (e.g., Crop, Season).
   • Handled missing values effectively.

2. Model Development

   • Chosen Model: Regression Forest for its accuracy and interpretability.
   • Metrics: High R², low MAE and MSE validated the model's reliability.

3. Feature Engineering

   • Combined features to model interactions.
   • Incorporated regional and seasonal trends.

---

Machine Learning Models Used

1. RandomForestClassifier:

   • Used for Seasonal Crop Yield Comparison (Classification Task).
   • Accuracy: 51.65%.

2. RandomForestRegressor:

   • Used for Regression Tasks:
     • Rainfall Impact on Yield Prediction:
       • MSE: 25924.09, R²: 96.76%.
     • Optimal Fertilizer Prediction:
       • MSE: 1.89e+14, R²: 97.43%.
     • State-based Yield Prediction:
       • MSE: 74883.98, R²: 90.65%.

Summary

• The Random Forest algorithm was chosen for its robustness, ability to handle non-linear relationships, and support for both classification and regression tasks.
• All trained models have been saved in a single file: all_models.pkl.

Installation

Prerequisites

• Python 3.8+
• Required libraries: pandas, numpy, scikit-learn, matplotlib, seaborn, tkinter.

Steps

1. Clone the repository:

   git clone https://github.com/Harshad071/AIML-Project.git
   cd AIML-Project

Code and Notebooks

Feature Engineering

The repository includes a Jupyter Notebook for Feature Engineering, which performs the following:

• Handles missing values using appropriate strategies for numerical and categorical columns.
• Encodes categorical variables using LabelEncoder.
• Scales numerical features using StandardScaler for improved model performance.
• Saves the cleaned and preprocessed data for further analysis.

File: feature_engineering.ipynb

Model Training

The Model Training notebook demonstrates:

• Training multiple models for various prediction tasks, including classification and regression.
• Tasks covered:
  1. Seasonal crop yield comparison (classification).
  2. Rainfall impact on yield prediction (regression).
  3. Optimal fertilizer and pesticide recommendation (regression).
  4. State-based yield prediction (regression).
• Saving all trained models in a single .pkl file for easy reuse.

File: model_training.ipynb

Tip: To replicate the results:

1. Ensure the dataset file crop_yield.csv is placed in the correct directory.
2. Run the notebooks in sequence:
   • Start with feature_engineering.ipynb to preprocess the data.
   • Proceed with model_training.ipynb to train and save the models.
3. Use the saved models (all_models.pkl) for predictions in downstream tasks.

Visualization

The notebooks also include:

• Histograms, boxplots, and count plots to explore data distributions and detect outliers.
• A correlation heatmap to analyze relationships between numerical features.

These visualizations provide valuable insights into the dataset and inform feature engineering decisions.