Tomorrows guest predictions as a strategic source
Project Profile
The prediction of standardized statistical key figures in tourism is always a challenge. Especially in times of COVID! We used an approach based on mobile phone data in Vienna and an approach using existing data sources of Statistik Austria like an extensive set of additive data to train our models to achieve a 3 month forecast for arrivals and overnight stays in Vienna. In cooperation with the AIT (Austrian Institute for Technology / http://www.ait.ac.at) we succeeded in producing an assessable forecast.
At a glance - essential project data
| Duration | From 2/1/2020 to 10/28/2020 with about 9 months of full engagement | |
|---|---|---|
| Data and Tools | Market - Tourism Sources • Mobile data Flight data VIE • EU-COVID Data • 30 years Arrivals/Nights • Weather • Google Trends • Global Holiday Calendar | |
| Integration | • Forecast information on client's systems • PowerBI as visualization of forecast data • Continuous process of data preparation with partner for our models | |
| AI Methods • Deep Learning • Time Series • ML - Regression |
Engagement Use-Case
Forecast model for arrivals and overnight stays for a city, region, ... over the period of 2 years forecast horizon.
Client motivation / Solution aims
Bugetation
To properly plan and deploy marketing efforts depending on future tourism currents
Achieve strategic tool for future campaigns
Get efficiency in tourism planning
AI Approach
| AI key technology used in our solution | Collaboration with AIT as independent quality assurance. Increase granularity of input data (overnight stays) by extrapolation with daily mobile data. | |
|---|---|---|
| Solution Approach | • Short-term model: LSTM TimeSeries • Long-term model: Logistic Growth (Gompertz) • DNN Timeseries model. Decomposition of the time series into three additive components for trend seasonality and residuals. | |
| Project Approach | Simply agile | |
| Project Type | Project | |
| ML Integration and ML Operations | • Operations Integration API • FTP • VIsualizationPower-BI • Azure Synapse |
Insights and Details
We use different approaches for forecasting in different stages of tourism development after lockdown
The DL model learns from the April data that a reduction is due and applies this also with COVID consideration - See left half of the picture!
Our forecast model as an interaction of many competences. From the extrapolation to the forecast!
The forecast results were presented in a small PowerBI dataset.
