FertiliserAdvisor

Fertilizer Recommendation System

Overview

This model predicts the fertilizer requirements for various crops based on input features such as crop type, target yield, field size, and soil properties. It utilizes a combination of Random Forest Regressor and Random Forest Classifier to predict both numerical values (e.g., nutrient needs) and categorical values (e.g., fertilizer application instructions).

Training Data

The model was trained on a custom dataset containing the following features:

• Crop Name
• Target Yield
• Field Size
• pH (water)
• Organic Carbon
• Total Nitrogen
• Phosphorus (M3)
• Potassium (exch.)
• Soil moisture

The target variables include:

Numerical Targets:

• Nitrogen (N) Need
• Phosphorus (P2O5) Need
• Potassium (K2O) Need
• Organic Matter Need
• Lime Need
• Lime Application - Requirement
• Organic Matter Application - Requirement
• 1st Application - Requirement (1)
• 1st Application - Requirement (2)
• 2nd Application - Requirement (1)

Categorical Targets:

• Lime Application - Instruction
• Lime Application
• Organic Matter Application - Instruction
• Organic Matter Application
• 1st Application
• 1st Application - Type fertilizer (1)
• 1st Application - Type fertilizer (2)
• 2nd Application
• 2nd Application - Type fertilizer (1)

Model Training

The model was trained using the following steps:

1. Data Preprocessing:

   • Handling missing values
   • Scaling numerical features using StandardScaler
   • One-hot encoding categorical features

2. Modeling:

   • Splitting the dataset into training and testing sets
   • Training a RandomForestRegressor for numerical targets using a MultiOutputRegressor
   • Training a RandomForestClassifier for categorical targets using a MultiOutputClassifier

3. Evaluation:

   • Evaluating the models using the test set with metrics like Mean Squared Error (MSE), Mean Absolute Error (MAE), and R-squared (R2) Score for regression, and accuracy for classification.

Evaluation Metrics

The model was evaluated using the following metrics:

• Mean Squared Error (MSE)
• Mean Absolute Error (MAE)
• R-squared (R2) Score
• Accuracy for categorical targets

How to Use

Input Format

The model expects input data in JSON format with the following fields:

• "Crop Name": String
• "Target Yield": Numeric
• "Field Size": Numeric
• "pH (water)": Numeric
• "Organic Carbon": Numeric
• "Total Nitrogen": Numeric
• "Phosphorus (M3)": Numeric
• "Potassium (exch.)": Numeric
• "Soil moisture": Numeric

Preprocessing Steps

1. Load your input data.
2. Ensure all required fields are present and in the expected format.
3. Handle any missing values if necessary.
4. Scale numerical features based on the training data.
5. One-hot encode categorical features (if applicable).

Inference Procedure

from huggingface_hub import hf_hub_download
import pandas as pd
from joblib import load
import numpy as np
from sklearn.preprocessing import LabelEncoder
from googletrans import Translator

# Initialize translator
translator = Translator()

# Download models from Hugging Face Hub
preprocessor_path = hf_hub_download(repo_id='your-username/your-repo', filename='preprocessor.joblib')
numerical_model_path = hf_hub_download(repo_id='your-username/your-repo', filename='numerical_model.joblib')
categorical_model_path = hf_hub_download(repo_id='your-username/your-repo', filename='categorical_model.joblib')

# Load the preprocessor and trained models
preprocessor = load(preprocessor_path)
numerical_model = load(numerical_model_path)
categorical_model = load(categorical_model_path)

# Define categorical targets (same as used during training)
categorical_targets = [
    'Lime Application - Instruction',
    'Lime Application',
    'Organic Matter Application - Instruction',
    'Organic Matter Application',
    '1st Application',
    '1st Application - Type fertilizer (1)',
    '1st Application - Type fertilizer (2)',
    '2nd Application',
    '2nd Application - Type fertilizer (1)'
]

# Example input data
new_data = {
    'Crop Name': 'maize(corn)',
    'Target Yield': 3600.0,
    'Field Size': 1.0,
    'pH (water)': 6.1,
    'Organic Carbon': 11.4,
    'Total Nitrogen': 1.1,
    'Phosphorus (M3)': 1.8,
    'Potassium (exch.)': 3.0,
    'Soil moisture': 20.0
}

# Preprocess the input data
input_df = pd.DataFrame([new_data])
input_transformed = preprocessor.transform(input_df)

# Make numerical predictions
numerical_predictions = numerical_model.predict(input_transformed)

# Make categorical predictions
categorical_predictions = categorical_model.predict(input_transformed)

# Load label encoders from Hugging Face Hub (if they are saved separately)
label_encoders = {col: load(hf_hub_download(repo_id='your-username/your-repo', filename=f'label_encoder_{col}.joblib')) for col in categorical_targets}

# Decode categorical predictions
categorical_predictions_decoded = {}
for i, col in enumerate(categorical_targets):
    le = label_encoders[col]
    try:
        decoded_labels = le.inverse_transform(categorical_predictions[:, i])
        # Translate to English
        translated_labels = [translator.translate(label, dest='en').text for label in decoded_labels]
        categorical_predictions_decoded[col] = translated_labels
    except ValueError as e:
        print(f"Error decoding predictions for {col}: {e}")
        categorical_predictions_decoded[col] = ["Unknown"] * len(categorical_predictions[:, i])

# Define numerical targets (same as used during training)
numerical_targets = [
    'Nitrogen (N) Need',
    'Phosphorus (P2O5) Need',
    'Potassium (K2O) Need',
    'Organic Matter Need',
    'Lime Need',
    'Lime Application - Requirement',
    'Organic Matter Application - Requirement',
    '1st Application - Requirement (1)',
    '1st Application - Requirement (2)',
    '2nd Application - Requirement (1)'
]

# Combine predictions into a single dictionary
predictions_combined = {**{col: numerical_predictions[0, i] for i, col in enumerate(numerical_targets)}, **categorical_predictions_decoded}

print("Predicted Fertilizer Requirements:")
for col, pred_value in predictions_combined.items():
    print(f"{col}: {pred_value}")