This group project was completed in a group of 5 as part of course 42577 Introduction to Business Analytics, at DTU in Fall 2023.
Pandas and scikit-learn packages are used to prepare data frames and train different models using the 2018 data from Citi Bike (New York), which contained over 17 million record entries. The notebook is divided into 4 sections:
- Section 1: Data Preprocessing & Visualization
- Section 2: Prediction Challenge
- Section 3: Exploratory Component
- Section 4: Conclusion
My main contributions are in sections 1 and 2.
The data is cleaned by removing improbable outliers, missing values, and rectifying the data types. A few visualizations are illustrated to identify patterns in the data.
The stations are placed into geographical clusters. Models are built to predict the demand for bike usage over the next 24 hours for a specific cluster of stations. Bike demand includes pick ups, drop offs, and consequently shortages in bikes available in these clusters.
A baseline model, linear regression model, and gradient boosted regressor model are used for comparison in performance.
The effect of seasonal and weather factors on the total and average bike trip durations is analyzsed using daily weather data in New York City extracted from Weather Data Services by Visual Crossing.