Project Overview

Objective

The project aimed to develop a time-series classification model to predict daily demand levels for Citi Bike rentals in Jersey City, NY, and classify them as High, Medium, or Low. This demand forecast helps optimize bike fleet distribution, improve resource utilization, and enhance customer satisfaction by ensuring bikes are available where and when needed.

Context

Citi Bike is one of the largest bike-sharing programs in the U.S., with demand patterns highly influenced by weather, rider type, and seasonal behavior. Jersey City, being a high-density area, often faces bike availability issues during rush hours or adverse weather. Predicting demand accurately can help tackle these challenges through smarter planning and dispatch.

Duration

The analysis was conducted over several weeks, involving phases of data preparation, exploration, modeling, and strategy development.

Tools and Methodologies

• Python (Pandas, NumPy, Matplotlib, Seaborn, Scikit-learn, PyTorch)
• EDA & Feature Engineering: Ride behavior + weather integration
• Modeling: Multinomial Logistic Regression and GRU-based Recurrent Neural Networks
• Evaluation: Permutation-based feature importance

The Approach and Process

Data Collection

Two datasets were collected covering the period from Jan 2022 to Sep 2023:

• 📈 Citi Bike Ride Data: Ride metadata including location, time, and rider type
• 🌦 Weather Data: Daily precipitation, snow, and temperature from climate APIs

These were merged on the daily level for time-series modeling.

Data Preprocessing

• Merged ride data with weather data by date.