Uber Trips Analysis

A geospatial clustering project exploring how repeated trip endpoints can be grouped into frequent user locations using DBSCAN and a K-Means baseline.

This repository focuses on per-user location clustering from ride data and is designed as a practical data-analysis project rather than a production recommendation system.

Problem

Riders often repeat the same places over time. This project tests whether historical trip endpoints can be grouped into recurring locations without needing to predefine the number of clusters.

Approach

The pipeline performs the following steps:

1. Load trip data and normalize column names
2. Validate numeric coordinates and timestamps
3. Remove invalid or out-of-range rows
4. Restrict analysis to the Lima bounding box
5. Stack start and end coordinates for each user
6. Run DBSCAN with haversine distance
7. Run a fixed-k K-Means baseline for comparison
8. Output per-user clustering summaries

Tech Stack

• Python
• pandas
• NumPy
• scikit-learn
• Plotly
• Jupyter-style exploratory workflow

Project Structure

src/
  data_cleaning.py
  clustering.py
  visualization.py
run_analysis.py
requirements.txt
tests/
  test_pipeline.py
data/
  sample_raw.csv

Running the project

Create a virtual environment and install dependencies:

python -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt

Run the sample pipeline:

python run_analysis.py --input data/sample_raw.csv --output-dir outputs/

Run tests:

pytest -q

Notes

• This repository does not include the original Uber Peru dataset.
• No ground-truth labels are provided for “correct” locations.
• DBSCAN is used as the primary method because it handles noise and does not require a predetermined cluster count.
• K-Means is included only as a comparison baseline and is not treated as the more correct method.

Limitations

• The project is limited to a single city dataset.
• 100m clustering can merge nearby places or split noisy points.
• A cluster is a repeated location pattern, not automatically a named place such as home or office.
• This project is exploratory and not a production recommendation service.