/cab_data_analysis

Passenger Prediction for cab drivers in San Francisco

Primary LanguageJupyter Notebook

Cab Spotting - CO2 Reduction Potential and Passenger Prediction

NbViewer Links:

  1. EDA - Temporal Heatmaps | 2. CO2 reduction | 3. Passenger Prediction | 4. Cab Clusters

Description

The repo contains four notebooks:

  1. 00_EDA_preprocessing.ipynb: The notebook states the business objectives and contains basic EDA and preprocessing of dataset.

    1. Notebook plots interactive temporal heatmaps to visualize cabtrails throughout the time period images/emptyvsfilled.png

    2. Notebook plots fraction of empty cabs images/frac_empty_cabs.png

  2. 01_CO2_reduction.ipynb: The notebook explores the first task of estimating yearly CO2 emission reduction caused due to unoccupied cabs.

    1. The notebook explores how to calculate the distance (vectorized haversine) between two coordinates
    2. Does more advanced EDA with trip duration and distances, ultimately finds and removes outliers
    • remove cabs with very high (or low) speeds
    1. States assumptions made and calculates the potential CO2 estimates (wheel and well-to-wheel estimates). images/co2

  3. 02_passenger_prediction.ipynb: Notebook takes the processed data and builds a model for taxi drivers to predict location of next passengers.

    1. Feature Engineering

      • time-based, location based and previous trip based features

    2. Very fast hist gradient boosting model, which can even be used for online training to do live predictions!

    3. Prediction example

      images/pred_ex

    4. MLOps demonstration

      • weights & biases dashboard with

  4. 03_cab_clusters.ipynb: This notebook attempts to find cab clusters using ML and domain knowledge.

    1. Uses advanced geoplotting to visualize the solution (notebook 0.)
    2. Segregates profit-making, under-utilized, efficient and liability cab drivers