How 0011ashwin/advanced-crop-yield-prediction-system works
Project type and purpose
• This is a full-stack, data-centric AI platform with both a web interface (Streamlit) and a mobile interface (Kivy/KivyMD).
• Purpose: ingest agricultural CSV/geospatial data, run exploratory analyses, train/serve ML models (yield prediction), integrate with Google’s Gemini AI for advanced Q&A and image analysis, and deliver results via interactive dashboards or a mobile app.
File structure
• Top-level:
• app.py – Streamlit entry point
• data/ – raw CSV/JSON datasets
• models/ – serialized model artifacts & metrics
• static/ – images for encyclopedia
• src/ – modular code organized into components, data loaders, model training, utils, and visualizations
• Notebook/ – Jupyter explorations
• src/components: high-level feature modules (data_exploration, feature_analysis, crop_information, model_training, yield_prediction, gemini_ai)
• src/data: data_loader
• src/models: model_trainer
• src/utils: auth, config, preprocessing, reporting
• src/visualizations: plot generators
Key insights from README
• Two user interfaces: web (Streamlit) and Android (Kivy/KivyMD, offline mode)
• Core ML tech: Scikit-learn, XGBoost/LightGBM, TensorFlow/PyTorch
• Visualization: Matplotlib, Seaborn, Plotly, GeoPandas
• External integration: Google Generative AI (Gemini API)
a. Main components
• Clients: Web Browser (interacts with Streamlit) & Android App (Kivy).
• Web Server: Streamlit application (app.py orchestrates feature modules).
• Mobile Runtime: Packaged Python/Kivy runtime on Android.
• Business Logic Modules (src/components):
– Data Exploration & Feature Analysis
– Crop Encyclopedia (crop_information)
– Model Training & Yield Prediction
– AI Assistant (gemini_ai)
• Data Layer:
– Local CSV/JSON files (data/)
– Model artifacts & metrics (models/)
• Visualization Layer (src/visualizations)
• Utility Layer (src/utils): config, auth, preprocessing, reporting
• External Services: Google Gemini AI API
b. Relationships & data flows
1. User (Browser/Mobile) → requests UI pages or mobile screens
2. UI → invokes backend modules via Streamlit/Kivy controllers
3. Data Loader reads CSV/JSON → Preprocessing → feeds Exploratory & Feature modules
4. Feature Analysis → Visualization → renders plots in UI
5. Model Training → persists model artifacts → updates metrics
6. Yield Prediction module loads model → predicts → results → UI
7. AI Assistant module → sends user queries/images to Gemini API → receives insights → UI
8. Reporting util generates downloadable reports from analysis results
c. Architectural patterns & design principles
• Modular Monolith: logically separated components but deployed as a single codebase
• Layered Pattern:
• Presentation (Streamlit/Kivy)
• Application (components orchestrating workflows)
• Domain (ML models, data logic)
• Infrastructure (file system, external API)
• Separation of Concerns: data loading, preprocessing, modeling, visualization, and integration each live in their own module
d. Relevant technologies/frameworks
• Presentation: Streamlit, Kivy/KivyMD
• Data Processing: Pandas, NumPy, GeoPandas
• Visualization: Matplotlib, Seaborn, Plotly
• ML: Scikit-learn, XGBoost, LightGBM, TensorFlow, PyTorch
• AI Integration: Google Generative AI (Gemini)
• Utils: Custom config/auth modules, reporting