Bike Demand Prediction Using Multiple Linear Regression

1. Introduction

BoomBikes, a U.S. bike-sharing provider, aims to forecast daily demand for shared bicycles in the aftermath of the COVID-19 lockdown. This project leverages historical usage and weather-related features to build a multiple linear regression model predicting total daily rentals (cnt). The insights will help BoomBikes optimize inventory, pricing, and station placement.

2. Problem Statement

• Goal: Identify significant factors influencing bike demand and quantify how well they explain rental counts.
• Target Variable: cnt (total number of daily bike rentals, including both casual and registered users).
• Key Questions:
  1. Which features significantly predict bike demand?
  2. How accurately can we model cnt using a linear regression approach?

3. Data Description

File	Description
bike_sharing.csv	Daily rental counts with features: season, yr, mnth, holiday, weekday, workingday, weathersit, temp, atemp, hum, windspeed, casual, registered, cnt
data_dictionary.csv	Definitions and coding for categorical variables (season, weathersit, yr, etc.)

4. Data Preparation

1. Load and inspect raw data (missing values, data types).
2. Convert ordinal codes to categories:
   • season: {1=Spring, 2=Summer, 3=Fall, 4=Winter}
   • weathersit: {1=Clear, 2=Mist, 3=Light Rain/Snow, 4=Heavy Rain/Snow}
3. Retain yr (0=2018, 1=2019) to capture yearly growth trend.
4. Drop casual and registered (components of cnt) to prevent leakage.
5. Train/test split: typically 70/30 or 80/20 based on mnth or random shuffle.

5. Model Building

1. Feature encoding: One-hot encode categorical variables (season, weathersit, weekday, etc.).
2. Standardization: Scale numerical predictors (temp, atemp, hum, windspeed) as needed.
3. Instantiate & train a LinearRegression model from scikit-learn.
4. Cross-validation (optional): Evaluate stability with k-fold CV.

6. Model Evaluation

• Generate predictions on the test set.
• Compute R-squared using:

  from sklearn.metrics import r2_score
  r2_score(y_test, y_pred)

• Analyze residuals for patterns (e.g., plot predicted vs. actual, residual histogram).

7. Environment & Setup

1. Clone the repo:

   git clone https://github.com/<username>/Bike-Demand-Regression.git
   cd Bike-Demand-Regression

2. Create virtual environment:

   python3 -m venv venv
   source venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

8. Usage

1. Launch Jupyter Notebook:

   jupyter notebook notebooks/bike_demand_lr.ipynb

2. Follow sections:
   • EDA and missing-value handling
   • Categorical conversion & feature engineering
   • Model training & evaluation
   • Residual analysis
3. View results and interpret model performance.

9. Future Work

• Explore nonlinear models (e.g., decision trees, GBMs).
• Incorporate external data (holiday calendars, special events).
• Build a dashboard for real-time demand forecasting.