A lightweight, explainable prototype that uses historical climate data (temperature, rainfall, humidity + two additional features) to classify climate risk and plant stress for tea gardens, and to generate actionable alerts and recommendations. Built as a Streamlit demo with simple, interpretable ML models (Random Forest / Decision Tree) and a rule-based decision engine.
▪ Climate stress level classification (Healthy / Mild / Severe) ▪ Risk-level classification (Low / Moderate / High / Extreme) ▪ Simple rule-based alert engine for actionable recommendations ▪ Streamlit dashboard with dataset upload, EDA, manual input sliders, prediction panel
tea-climate-analyzer
—> data
→ Indian_Climate_Dataset_2024_2025.csv
—> notebooks
→ eda.ipynb
—> src
→ features.py
→ modeling.py
→ app.py
—> models
→ rf_risk.pkl
→ dt_stress.pkl
—> requirements.txt
—> README.md
—> LICENSE
data/Indian_Climate_Dataset_2024_2025.csv
High-level pipeline: 1.Data ingestion and parsing (date handling) 2.Data cleaning (missing values, outlier handling) 3.Feature engineering (rolling aggregates, deviations, CSI) 4.Label creation (CSI → stress and risk labels) 5.Modeling: RandomForestClassifier for risk (baseline) DecisionTreeClassifier or LogisticRegression for stress (explainable) 6.Evaluation (accuracy, macro F1, confusion matrix, feature importances) 7.Model serialization (joblib.dump) and Streamlit-based delivery 8.Rule engine for alerts and recommendations
▪ Integrate live weather APIs for near real-time inputs and automated model retraining on incoming data (live weather retraining). ▪ Add remote-sensing signals (NDVI / satellite rainfall validation). ▪ Include soil moisture, leaf wetness sensors, or farm-level sensors for better ground truth. ▪ Build a retraining pipeline and scheduled evaluation for model drift.