The rapid growth of bike-sharing systems has led to the generation of large volumes of data related to weather conditions, temporal patterns, and user behavior. Accurately predicting bike rental demand is critical for:

• Optimizing bike distribution and resource allocation
• Improving customer satisfaction
• Reducing operational and maintenance costs

This project builds an end-to-end machine learning regression system to predict the total number of bike rentals (cnt) using historical data and multiple influencing factors.

Bike rental demand is affected by a combination of:

• 🌦 Weather conditions: temperature, humidity, windspeed
• 📆 Temporal patterns: season, month, weekday, working day
• 🎉 External factors: holidays, weather situations

The challenge is to design a robust regression-based ML pipeline that can effectively learn these patterns and deliver accurate predictions.

• Perform data preprocessing and exploratory data analysis (EDA)
• Train and compare multiple regression models
• Evaluate models using multiple performance metrics
• Track experiments, metrics, and models using MLflow
• Register and load the best-performing model
• Build an interactive Streamlit application for predictions and visualization

The following regression models are trained and evaluated:

• Linear Regression
• Random Forest Regressor
• Gradient Boosting Regressor
• K-Nearest Neighbors (KNN) Regressor
• Support Vector Regressor (SVR)
• LightGBM Regressor (LGBM)

Each model is evaluated using:

• R² Score
• Adjusted R²
• Mean Absolute Error (MAE)
• Mean Squared Error (MSE)
• Root Mean Squared Error (RMSE)
• Accuracy-like Metric → R² × 100

This project leverages MLflow to ensure reproducibility and experiment tracking:

• Logs model parameters
• Tracks evaluation metrics
• Stores trained models
• Registers the best-performing model
• Enables model loading for deployment

An interactive Streamlit web app is built to:

• Input environmental and temporal features
• Predict bike rental demand in real time
• Visualize predictions and model performance
• Load models directly from MLflow Registry

git clone https://github.com/your-username/bike-demand-prediction.git

mlflow ui

Identify the best-performing regression model for bike demand prediction

Understand feature importance impacting bike rentals

Enable reproducible ML workflows using MLflow

Python

Scikit-learn

LightGBM

MLflow

Streamlit

Pandas, NumPy, Matplotlib, Seaborn