🚲 Bike Sharing Demand Prediction — Linear Regression

This project analyzes hourly bike rental demand using the Bike Sharing Dataset (2011–2012).
It explores temporal patterns, weather influence, and environmental factors affecting bike usage, followed by building and evaluating a Linear Regression model as a baseline predictor.

The goal of this project is to:

• Understand how weather, season, time, and user behavior influence hourly bike demand
• Perform detailed EDA (temporal, seasonal, weather-based trends)
• Build a baseline Linear Regression model
• Interpret coefficients and feature importance
• Identify model strengths, weaknesses, and improvement paths

The regression model achieved:

• R² Score: 0.396
• MAE: 104.08
• RMSE: 138.80

Although linear regression provides interpretability and transparency, the dataset exhibits complex nonlinear patterns that require advanced models.

The dataset contains hourly bike rental records from a bike-sharing system, with 17 features including:

• Time-related: season, year, month, hour, weekday, workingday
• Weather-related: temperature (temp), feeling temperature (atemp), humidity (hum), windspeed
• User counts: casual, registered
• Target variable: cnt (total bike rentals per hour)

No missing values were found, and preprocessing included datetime merging, label mapping, and removal of redundant fields.

• Clear bimodal peaks at 8 AM and 5–6 PM, aligning with commuting times
• Weekends/holidays show flatter demand curves, indicating recreational use

• Spring has lowest demand
• Summer, Fall, Winter show similar higher demand levels

• Clear & Misty conditions → highest usage
• Light rain/snow → moderate drop
• Heavy rain/snow → very low usage

• temp and atemp are highly correlated (0.99)
• Humidity negatively impacts demand
• Registered users strongly influence total demand

• Baseline Multiple Linear Regression
• Predictors include hour, year, holiday, weekday, workingday, weather condition, temperature, humidity, windspeed, and season.

• Hour is the strongest predictor (commute-linked behavior)
• Temperature increases demand
• Humidity suppresses ridership
• Windspeed has negligible impact

• Linear model underfits during peak hours
• Produces negative predictions under low usage
• Residuals show non-linearity and heteroscedasticity

Advanced models (Random Forest, Gradient Boosting, XGBoost) would significantly improve accuracy.

• Bike demand is driven heavily by commuting time and favorable weather
• Humidity and bad weather drastically reduce ridership
• Demand patterns differ across working days and weekends
• Baseline linear regression explains ~40% of variability — good start but insufficient for real deployment

• Python
• NumPy, Pandas
• Matplotlib, Seaborn
• Scikit-learn
• Jupyter / Google Colab