Rainfall_Prediction_using_Machine_Learning

🌧️ Rainfall Prediction using Machine Learning – End-to-End Classification Project with Deployment

Deployed on Streamlit: View Live App

Python | Scikit-Learn | Logistic Regression | Gradient Boosting | XGBoost

---

Table of Contents

• Project Overview
• Dataset
• Data Exploration & Cleaning
• Data Preprocessing
• Model Selection & Evaluation
• Hyperparameter Tuning
• Final Model Performance
• Predictions
• Key Takeaways
• Technologies & Libraries Used

---

Project Overview

This project aims to predict rainfall (Yes/No) based on various meteorological features such as temperature, pressure, humidity, wind speed, and cloud cover.

The objective is to help in early weather forecasting and decision-making for agriculture and disaster management, by building an accurate ML model that learns rainfall patterns from past weather data.

---

Dataset

• Source: Helping Hand Kaggle Competition
• Rows: ~3,000
• Columns: 12 (11 features + 1 target)

Features:

• day – Day of year (1–365)
• pressure – Atmospheric pressure (hPa)
• maxtemp – Maximum temperature (°C)
• temparature – Average temperature (°C)
• mintemp – Minimum temperature (°C)
• dewpoint – Dew point temperature (°C)
• humidity – Relative humidity (%)
• cloud – Cloud cover (%)
• sunshine – Sunshine duration (hours)
• winddirection – Wind direction (°)
• windspeed – Wind speed (km/h)
• Target: rainfall (0 = No Rain, 1 = Rain)

---

Data Exploration & Cleaning

• Checked for missing/null values → handled or confirmed none.
• Verified valid ranges:
  • Pressure: 999–1035 hPa
  • Temperature: 4–36°C
  • Humidity: 39–98%
  • Wind Speed: 4–59 km/h
• Converted target (rainfall) into binary values.
• No duplicates or inconsistent entries found.

---

Data Preprocessing

• Split into train (85%) and test (15%) sets.
• Applied StandardScaler on all continuous features.
• Saved the scaler as scaler.pkl for deployment consistency.
• Used scaled datasets for model training and evaluation.

---

Model Selection & Evaluation

Models evaluated:

Model	Accuracy	Precision	Recall	F1-Score	ROC-AUC
Gradient Boost	0.881	0.884	0.967	0.924	0.904
Logistic Regression	0.878	0.889	0.955	0.921	0.922
XGBoost	0.875	0.875	0.971	0.921	0.884
Random Forest	0.872	0.883	0.955	0.918	0.898
AdaBoost	0.869	0.885	0.947	0.915	0.914
KNN	0.848	0.865	0.943	0.902	0.861
Decision Tree	0.815	0.868	0.886	0.877	0.746

Gradient Boosting Classifier and Logistic Regression were the top-performing models overall.

---

🔧 Hyperparameter Tuning

Applied GridSearchCV with 5-fold cross-validation.

Gradient Boosting Classifier

Best Parameters:

{
  'learning_rate': 0.01,
  'max_depth': 3,
  'n_estimators': 300,
  'subsample': 0.8
}

Logistic Regression

Best Parameters:

{
  'C': 0.1,
  'penalty': 'l2',
  'solver': 'liblinear'
}

---

Final Model Performance

After tuning, the Logistic Regression model was selected as the final model due to its excellent balance of accuracy, interpretability, and ROC-AUC score.

Metric	Logistic Regression (Final Model)
Accuracy	0.878
Precision	0.890
Recall	0.955
F1-Score	0.921
ROC-AUC	0.922

---

Predictions

In the deployed Streamlit App, users enter daily weather conditions:

• Day
• Pressure
• Max / Min / Avg Temperature
• Dew Point
• Humidity
• Cloud Cover
• Sunshine Hours
• Wind Direction
• Wind Speed

Inputs are scaled via scaler.pkl, and the trained Logistic Regression model (model.pkl) predicts whether it will rain (1) or not (0).

---

Key Takeaways

• Logistic Regression provided the most stable and interpretable predictions.
• Gradient Boosting achieved high recall and can be used for more sensitive rainfall detection.
• Humidity, dew point, and cloud cover strongly influence rainfall probability.
• Feature scaling improves model reliability across various data ranges.
• Deployment through Streamlit provides an intuitive interface for real-time rainfall prediction.

---

Technologies and Libraries Used

Programming Language: Python 3.10

Libraries:

• pandas – Data manipulation
• numpy – Numerical operations
• scikit-learn – Preprocessing, model selection, evaluation
• xgboost – Advanced boosting algorithm
• matplotlib, seaborn – Data visualization
• pickle – Model serialization
• streamlit – Web app deployment