https://github.com/csakig/bike-sharing-demand-analytics

Advanced analytics on Bike Sharing data using Random Forest, Gradient Boosting, and SARIMAX forecasting.
https://github.com/csakig/bike-sharing-demand-analytics

data-science machine-learning portfolio python scikit-learn time-series

Last synced: 2 months ago
JSON representation

Advanced analytics on Bike Sharing data using Random Forest, Gradient Boosting, and SARIMAX forecasting.

• Host: GitHub
• URL: https://github.com/csakig/bike-sharing-demand-analytics
• Owner: csakig
• License: mit
• Created: 2025-12-20T11:44:04.000Z (6 months ago)
• Default Branch: main
• Last Pushed: 2025-12-20T12:16:53.000Z (6 months ago)
• Last Synced: 2025-12-22T16:38:47.413Z (6 months ago)
• Topics: data-science, machine-learning, portfolio, python, scikit-learn, time-series
• Language: Jupyter Notebook
• Homepage:
• Size: 1.85 MB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# 🚲 Bike Sharing Demand Analytics

### 🚀 Optimizing Fleet Operations with Machine Learning & Time Series Forecasting

![Python](https://img.shields.io/badge/Python-3.8%2B-blue?style=for-the-badge&logo=python&logoColor=white)

![Scikit-Learn](https://img.shields.io/badge/scikit--learn-F7931E?style=for-the-badge&logo=scikit-learn&logoColor=white)

![Statsmodels](https://img.shields.io/badge/Statsmodels-Time_Series-blueviolet?style=for-the-badge)

![Status](https://img.shields.io/badge/Status-Completed-success?style=for-the-badge)

---

## 📸 Executive Summary

This project aims to solve a critical logistics problem: **predicting bike availability demand to optimize rebalancing operations.**

By analyzing historical usage patterns, I developed a machine learning pipeline that not only predicts exact rental counts ($R^2=0.94$) but also forecasts future demand trends using SARIMAX, enabling proactive fleet management.

### 🏆 Key Visual Insight

*The analysis revealed distinct "Commuter" vs. "Leisure" patterns, which became the primary feature for the model.*

![Demand by Hour](demand_by_hour.png)

*(Fig 1. Hourly demand clearly separates working days (orange) from weekends (blue), dictating the feature engineering strategy.)*

---

## 📊 Business Value & Recommendations

Based on the data, the following operational changes are recommended:

| Insight | Actionable Recommendation | Expected Impact |

| :--- | :--- | :--- |

| **Rush Hour Spikes** | Deploy rebalancing trucks at **9:00 AM** and **3:00 PM** (post-peak). | Prevent station "starvation" during commuting hours. |

| **Weather Sensitivity** | Push notifications/discounts when **Temp < 10°C** or **Humidity > 80%**. | Stabilize revenue during low-demand weather conditions. |

| **Seasonal Trends** | Schedule heavy maintenance in **Spring** (lowest demand). | Maximize fleet availability during Fall/Summer peaks. |

---

## 🧠 Technical Approach

### 1. Data Processing & EDA

* **Feature Engineering:** extracted `hour`, `day_of_week`, and `season` from timestamps.

* **Outlier Detection:** Removed anomalies in windspeed and humidity data.

* **Correlation Analysis:** Identified Temperature as the strongest driver of demand.

![Correlation / Feature Importance](feature_importance.png)

### 2. Modeling Strategy

I tested four algorithms to find the optimal balance between accuracy and interpretability.

| Model | Task | Performance Metric | Verdict |

| :--- | :--- | :--- | :--- |

| **Random Forest** | Regression | **$R^2 = 0.94$** | 🏆 **Best Performer** (Selected) |

| **Gradient Boosting** | Classification | **F1-Score = 0.92** | Excellent for Peak Detection |

| **Linear Regression** | Regression | $R^2 = 0.39$ | Underfitted (Non-linear data) |

### 3. Forecasting (Time Series)

Using **SARIMAX (Seasonal ARIMA with eXogenous variables)**, I modeled the weekly seasonality to project demand for the next 30 days.

![30 Day Forecast](forecast_30_days.png)

*(Fig 3. The model (red) successfully captures the weekly cycle and growth trend with a narrow 95% confidence interval.)*

---

## 🛠️ Installation & Usage

# 1. Clone the repo

git clone [https://github.com/csakig/bike-sharing-demand-analytics.git](https://github.com/csakig/bike-sharing-demand-analytics.git)

# 2. Install dependencies

pip install -r requirements.txt

# 3. Run the analysis

jupyter notebook Bike_Sharing_Demand_Analysis.ipynb

---

📂 Repository Structure

Bike_Sharing_Demand_Analysis.ipynb: Main analysis notebook.

data.csv: Historical dataset.

demand_by_hour.png: Visualization asset.

feature_importance.png: Visualization asset.

forecast_30_days.png: Visualization asset.

---

Author: csakig | Aspiring Data Scientist