Weather Classification

Weather Type Classification using SVM

A machine learning project that predicts weather type (Rainy, Sunny, Cloudy, Snowy) from meteorological features using Support Vector Machine (SVM) classifiers.

---

Dataset

Source: Weather Type Classification – Kaggle (Nikhil Narayan)
File: weather_classification_data.csv
Size: 13,200 rows × 11 columns

Feature	Type	Description
temperature	int	Temperature in °C
humidity	int	Humidity percentage
wind_speed	float	Wind speed in km/h
precipitation (%)	int	Precipitation percentage
cloud_cover	str	Cloud cover description
atmospheric_pressure	float	Atmospheric pressure in hPa
uv_index	int	UV index
season	str	Season of recording
visibility (km)	float	Visibility in kilometres
location	str	Type of location
weather_type	str	Target — Rainy / Sunny / Cloudy / Snowy

---

Project Structure

weather_classification_data.csv   # Dataset
weather_classification.ipynb      # Main notebook
README.md

---

Requirements

pip install pandas scikit-learn seaborn matplotlib

Python 3.8+ recommended.

---

Workflow

The notebook is organised into six tasks:

Task 1 — Data Preparation & Exploration

• Load the CSV into a pandas DataFrame
• Check shape, missing values, and data types
• Visualise key features:
  • season → pie chart
  • temperature, humidity, wind_speed → histograms
  • precipitation (%) → box plot

Task 2 — Data Transformation

• One-hot encode categorical columns: cloud_cover, location, season
• StandardScaler applied to all seven numerical features

Task 3 — SVM with Linear Kernel

• 70/30 train-test split (random_state=42)
• Train SVC(kernel='linear')
• Evaluate with accuracy score, classification report, and confusion matrix

Task 4 — SVM with RBF Kernel

• Train SVC(kernel='rbf') on the same split
• Compare accuracy and evaluation metrics against the linear kernel

Task 5 — Hyperparameter Experimentation

• Train a custom RBF SVM with C=0.5, gamma='auto', degree=2
• Observe the effect on accuracy and per-class metrics

Task 6 — Sklearn Pipeline

• Build a Pipeline([StandardScaler → SVC(rbf)])
• Fits and evaluates end-to-end in a single object — clean, reproducible, and leak-proof

---

How to Run

jupyter notebook weather_classification.ipynb

Run all cells in order. Each task section is self-contained with inline comments.

---

Key Concepts

• SVM (Support Vector Machine): Finds the optimal hyperplane that maximises the margin between classes.
• Linear kernel: Works well when classes are linearly separable.
• RBF kernel: Maps data to a higher-dimensional space; handles non-linear boundaries.
• Pipeline: Chains preprocessing and modelling into one estimator, preventing data leakage during cross-validation.

---

References

• sklearn.svm.SVC documentation
• Kaggle Dataset