Uber Ride Analysis

A data science project analyzing Uber ride data to uncover patterns in ride demand, fare pricing, and peak hours, including a fare prediction model.

Dataset

• Source: Uber Fares Dataset (Kaggle)
• Size: 200,000 rides (193,188 after cleaning)
• Features: pickup datetime, pickup/dropoff coordinates, fare amount, passenger count

Project Structure

uber-ride-analysis/
├── data/
│   ├── uber.csv              # Raw dataset
│   └── uber_cleaned.csv      # Cleaned dataset 
├── visuals/                  # All saved chart images
├── preprocessing.py          # Data loading, cleaning, feature engineering
├── eda_visuals.py            # 5 exploratory visualizations
├── fare_prediction.py        # Regression model for fare prediction
├── tests.py                  # Automated data validation tests
├── main.py                   # Runs the full pipeline
└── README.md

Data Preprocessing

• Removed unused ID columns (Unnamed: 0, key)
• Dropped rows with missing values
• Extracted hour, day, month, year, day_of_week from pickup datetime
• Calculated trip distance using the Haversine formula (great-circle distance from GPS coordinates)
• Filtered unrealistic values: fares outside $0–200, distances outside 0–100km, passenger counts outside 1–6

Visualizations

1. Bar plot — Rides by hour of day (peak hour identification)
2. Line plot — Average fare trend across months
3. Scatter plot — Trip distance vs fare amount
4. Violin plot — Fare distribution across passenger counts
5. Heatmap — Ride frequency by day of week vs hour

Predictive Model

A Linear Regression model predicts fare_amount using distance, hour, passenger count, month, and day.

• Mean Absolute Error: $2.39
• R² Score: 0.725
• Most influential feature: trip distance (coefficient 2.19), far outweighing passenger count, month, hour, or day

Key Insights

• Peak ride hours are 6 PM – 10 PM on weekdays
• Weekend early mornings (12 AM – 3 AM) show unusually high demand compared to weekday early mornings
• Fare is driven almost entirely by distance, not passenger count
• The model performs well for typical short-to-mid trips but underestimates a small number of rare, high-fare outlier trips (likely long-distance/airport rides)

How to Run

pip install pandas numpy matplotlib seaborn scikit-learn
python main.py

Run individual modules directly (python eda_visuals.py, python fare_prediction.py) to see chart pop-ups and model output interactively.

Tests

python tests.py

Validates: no negative fares, no negative distances, passenger count in valid range, no missing values.