Bike-Sharing System Analysis and Forecasting

Overview

This repository provides an end-to-end analysis of a bike-sharing system, focusing on demand forecasting and trip duration analysis. By leveraging advanced machine learning models, it offers actionable insights to optimize operations and enhance customer experience.

Dataset

The analysis is based on two datasets:

1. Demand Data:
   • Features: Date, Hour, Demand, Weather Conditions, Public Holiday, Working Day.
   • Objective: Forecast hourly bike demand.
2. Duration Data:
   • Features: Duration, Date, Borough, Weather Conditions, Public Holiday, Working Day.
   • Objective: Analyze trip durations and identify key factors affecting them.

Objective

1. Demand Forecasting:
   • Predict hourly bike demand to improve resource allocation.
2. Trip Duration Analysis:
   • Identify factors affecting trip durations and suggest operational improvements.

Methodology

1. Data Preprocessing:

   • Cleaned and prepared data for analysis.
   • Engineered features like weather conditions, public holidays, and weekdays.

2. Demand Forecasting:

   • Models Used: ARIMA, SARIMA, and LSTM.
   • Evaluated model performance using R² and RMSE metrics.
   • Forecasted demand for future months.

3. Duration Analysis:

   • Models Used: Random Forest, Gradient Boosting, and Neural Networks.
   • Analyzed feature importance to identify key factors affecting durations.
   • Clustering and segmentation for borough-level insights.

4. Performance Evaluation:

   • Compared models across multiple train-test splits (70:30, 60:40, 80:20).
   • Captured metrics (MAE, RMSE, R²) and saved predictions for further analysis.

Future Work

• Incorporate Real-Time Data:
  • Use live weather and event data to improve forecasting accuracy.
• Advanced Models:
  • Experiment with Transformer-based models and NeuralProphet for forecasting.
• Integration:
  • Build a dashboard for real-time visualization and decision-making.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Disclaimer

This repository is the proprietary work of Kayalks and is part of a Master's dissertation project.

Licensing

• All Rights Reserved: Unauthorized use, distribution, or reproduction of this repository is strictly prohibited.
• For permission to use this work, please contact Kayalvizhi Selvaraj.