forecasting_prophet

People count Forecasting ✨

We developed with Prophet a forecasting model to predict the number of people waiting at the stops in a future moment based on the past recorded data. In the system, there will be a Prophet-object for each stop. Since it is just a demo project the recorded data was generated using a script, so the model is not accurate.

Our workflow was:

1. fake data generation -> data_generation.py
2. model training -> model_train_test.py
3. model prediction -> model_predict.py

model_plot.py is the same of model_train_test.py, but it shows some plots about Prophet's predictions.

1. Fake data generation

data_generation.py generates time series (day / number of peoples that have booked the stop) needed to train and test the Prophet model. Data are generated using a composition of a gaussian distribution and an ad-hoc distribution in order to resemble the real-world peak hours of traffic (we assume at 12:00 and 19:00). Data are saved in a .csv file using the format timestamp, number of people:

• training data are saved in the file train_data.csv
• test data are saved in the file test_data.csv

The script expects the initial and final dates of data in the format YYYY-MM-DD hh:mm:ss. A running example:

python data_generation.py

Acquisizione intervalli temporali di training e testing...
Inserire i datetime richiesti nel formato: 'YYYY-MM-DD hh:mm:ss'
Data di inizio training:2022-01-01 06:00:00
Data di fine training:2022-12-31 21:30:00
Data di inizio testing:2023-01-01 06:00:00
Data di fine testing:2023-01-31 21:30:00
Generazione dati di training...
Generazione dati di testing...
Fatto!

2. Model training

model_train_test.py trains the model and tests it. This script expects training and testing data in .csv format.:

python model_train_test.py train_data.csv test_data.csv

The trained model is saved in the file serialized_model.json.

Some example charts (if you run model_plot.py):

3. Model prediction

model_predict.py makes predictions. The script expects a train model file (json-serialized) and an output file in csv format (timestamp, number of people):

python model_predict.py model_file.json data_file.csv

Acknowledgements

• Official Prophet page -> https://facebook.github.io/prophet/
• Official Github Prophet page -> https://github.com/facebook/prophet
• Useful quickstart guide to Prophet -> https://www.youtube.com/watch?v=j0eioK5edqg