🌾 Crop Yield Prediction

📋 Project Overview

An end-to-end machine learning system that predicts crop yield based on environmental conditions, soil properties, and farm management practices. The system integrates multiple datasets, performs intelligent feature engineering, and serves predictions via a REST API.

ML Pipeline Workflow

📈 Yield Distribution Analysis

Distribution of crop yields across the unified dataset shows a right-skewed pattern, with most yields concentrated between 200-800 kg/hectare.

Figure 1: Histogram showing the distribution of crop yields. Mean yield: ~636 kg/ha | Median: ~580 kg/ha

🔗 Feature Correlations

Correlation analysis reveals key relationships between features and yield. Fertilizer amount and irrigation schedule show strongest positive correlations.

Figure 2: Correlation heatmap of all numerical features. Darker colors indicate stronger correlations.

🌾 Yield by Crop Type

Comparison of average yields across different crop types reveals significant variations. Sugarcane shows highest yields while pulses show lowest.

Figure 3: Box plot comparing yield distributions across crop types. Sugarcane leads with highest average yield.

📍 Yield by State

Regional analysis shows yield variations across Indian states. Punjab and Haryana show consistently higher yields due to better irrigation infrastructure.

Figure 4: Average crop yields by state. Northern states show higher yields on average.

🌤️ Weather & Soil Impact

Scatter plots reveal the relationship between key environmental factors and crop yield. Optimal ranges exist for each parameter.

Figure 5: Multi-panel scatter plots showing how weather and soil parameters influence crop yield.

⭐ Feature Importance

Gradient Boosting model reveals the relative importance of each feature in predicting yield. Fertilizer amount and rainfall are the most influential predictors.

Figure 6: Feature importance scores from the Gradient Boosting model. Top features: Fertilizer, Rainfall, Temperature.

🎯 Model Prediction Analysis

Visualization of actual vs predicted yields shows the model's accuracy across different yield ranges. The tight clustering around the diagonal indicates strong predictive performance.

Figure 7: Actual vs Predicted yield scatter plot with regression line. R² = 0.9627 indicates excellent model fit.

🔍 Data Quality & Outlier Analysis

Before generating the synthetic dataset, the original data was analyzed for outliers. Issues like temperature values in Fahrenheit (>50°C) were identified and addressed.

Figure 8: Outlier detection analysis showing data distribution and identified anomalies in the original dataset.

📊 Synthetic Dataset Info

A synthetic dataset of 75,000 records was generated to ensure complete data coverage with realistic correlations between agricultural features.

Data Coverage - All Features at 100%

🤖 Model Comparison

Four regression algorithms were trained and evaluated using 5-fold cross-validation. Gradient Boosting achieved the best overall performance and was selected for production.

🛠️ Technology Stack

API Endpoints

✅ Summary & Next Steps