Crowd count prediction and dashboard

This application predicts the 7-day crowdedness of 3 locations in the city centre of Amsterdam, to keep the citty safe. It gathers data from crowd-count sensors, weather stations, nearby (public) transport vendors and more, processes and cleans these to be used with LSTM and XGBoost models. This information along with valuable statistics is served to the user through an interactive dashboard.

Our predictions were a 68% F1 accuracy improvement to the baseline due to improved models and data blending!
All crowd-count data has been replaced with dummy data for privacy reasons.

You can check the paper here.

---

Data Pipeline

---

Deploying

Setting up the environment

Python venv

Make sure virtualenv is installed on your machine. Then do these following steps:

• virtualenv venv -p python3.7

• source venv/bin/activate

• pip install -r requirements.txt

conda venv

• conda create --name <envname> python=3.7
• conda activate <envname>
• pip install -r requirements.txt
• Optionally, if you run on Mac, run the following:

brew install libomp
conda install -c conda-forge py-xgboost

Go to src directory, run this from the shell/terminal:

• streamlit run app.py

You should see the app pop up.