Amsterdam Metro Crowdedness Prediction

The aim of this full-stack project is to predict and visualize crowdedness for 1 week ahead in 3 metro stations of Amsterdam: Centraal Station, Station Zuid and Station Bijlmer ArenA. Except for the number of check-ins & check-outs for each station, external factors are considered such as weather, events, holidays, vacations and COVID-19 pandemic.

Description

The project consists of the following components:

• instagram-event-scraper → scraper for events from Instagram using instagram's public URLs
• ticketmaster-event-fetcher → fetcher for events from Ticketmaster API
• model → back-end and front-end for making predictions
  • data_utils.py → helper functions for data manipulation and logging
  • model_utils.py → functions for model pipeline
  • predictions.ipynb → notebook for running model pipeline
  • predictions_server.py → Flask server for running model pipeline
  • UI → front-end for running model pipeline

Model Pipeline

1. Read and preprocess data
2. Merge data of external factors (e.g. weather) with check-ins & check-outs per hour
3. Interpolate missing check-ins & check-outs by using Random Forest algorithm
4. Split dataset into training, validation and test set
5. Create a separate Random Forest model for each of the 3 metro stations
6. Train each model with historical data (X)
7. Predict the check-ins & check-outs for each hour for 1 week ahead (Y)

Getting Started

Dependencies

• Python 3.7+
• All the libraries included in requirements.txt

Installing

• Run pip install -r requirements.txt
• Datasets for check-ins & check-outs (model/data/gvb/ & model/data/gvb-herkomst/), ** weather** (model/data/knmi/) and events (model/data/events/) are expected to be in model/ as per this directory structure:

model
└───data
    └───gvb
    │   └───<year>
    │   │   └───<month_number>
    │   │   │   └───<day_number>
    │   │   │       │   <csv_or_json.gz>
    │   │   │       │   ...
    │   │   │
    │   │   └───...
    │   └───...
    └───gvb-herkomst
    │   └───<year>
    │   │   └───<month_number>
    │   │   │   └───<day_number>
    │   │   │       │   <csv_or_json.gz>
    │   │   │       │   ...
    │   │   │
    │   │   └───...
    │   └───...
    └───knmi
    │   └───knmi
    │   │   └───<year>
    │   │   │   └───<month_number>
    │   │   │   │   └───<day_number>
    │   │   │   │       │   <json>
    │   │   │   │       │   ...
    │   │   │   │
    │   │   │   └───...
    │   │   └───...
    │   └───knmi-observations
    │       └───<year>
    │       │   └───<month_number>
    │       │   │   └───<day_number>
    │       │   │       │   <json>
    │       │   │       │   ...
    │       │   │
    │       │   └───...
    │       └───...
    └───events
        │   events_zuidoost.xlsx
        │
        └───instagram
        │   │   <csv>
        │   │   ...
        │
        └───ticketmaster
            │   <csv>
            │   ...

• WARNING: For the model to produce valid predictions, check-ins & check-outs (model/data/gvb/ & model/data/gvb-herkomst/) and weather data (model/data/knmi/) should be manually up-to-date

Executing programs

instagram-event-scraper

• Modify usernames array in scraper.py to include the usernames of the accounts which you want to be scraped
• Go to instagram-event-scraper/ and run python scraper.py
• After execution, instagram-event-scraper/events.csv will be updated with the scraped events

ticketmaster-event-fetcher

• Create ticketmaster-event-fetcher/config.py containing api_key=EXAMPLE where EXAMPLE is a placeholder for your Ticketmaster API key
• Modify year_to_fetch variable in fetcher.py to fetch events for the year of your choice
• Go to ticketmaster-event-fetcher/ and run python fetcher.py
• After execution, a file with format ticketmaster-event-fetcher/events_amsterdam_center_DATE_TIME_UTC.csv will be created with the fetched events

model

• Using model/predictions.ipynb:
  • Modify config.ini for the model to use the feature configuration of your choice
  • Run model/predictions.ipynb
  • See below bullet point "After execution"
• Using front-end and back-end server:
  • Go to model/, run python predictions_server.py and wait for the server output to show "Preprocessing finished" and be up
  • Go to model/UI/, run python test.py and wait for the front-end server to be up
  • Open the URL of the front-end server on a browser
  • Choose your desired parameters for the model and press "Submit"
  • After execution
    • If you press to any of the 3 available metro stations in the map, the graph should be updated with the current predictions
    • Each station's folder in model/output/ will be updated with a new file with format prediction_next_week_CURRENT-DATE.csv which will contain the current predictions
    • NOTE: Only if you ran the model using model/predictions.ipynb notebook, then model/output/models_log.csv will be updated with the model's parameters and metrics

Acknowledgments

• instagram-scraper