Challenge provided by Cascais Municipality

Predicting air quality for outdoor activities

Predicting the pollution levels in the future could help regulate traffic, reduce the air quality health impact, or create targeted marketing campaigns to raise awareness among the population.

According to the WHO, there are several health consequences of poor air quality, such as an increased risk of respiratory infections, heart disease, and lung cancer. While reducing air pollution is a slow process, small improvements can already impact the health of many citizens. This challenge focuses on predicting the ideal locations for outdoor activities, considering the air quality in the city of Cascais.

Since 2020, Cascais has been implementing an amount of air quality sensors. This air quality system aims to monitor what’s happening in Cascais and identify areas where there’s a need to act and improve air quality while improving and creating better experiences for its inhabitants.


Create a model that predicts the best locations for outdoor activities by minimizing the effect of air pollution.

United Nations SDG 

GOAL 11: Sustainable Cities and Communities


The following datasets were provided to the participants:

  • Air quality measurements from 11 sensors with daily averages provided by Cascais Municipality.
  • Points of interest for outdoor activities provided by Cascais Municipality.


One team used the weather information from OpenWeatherMap to complete the information missing from what was provided by the city of Cascais. They also added extra information regarding road traffic within a 3km radius of the sensor. It was suggested to include UV measurements in the future as it might help predict ozone levels

Methods and Techniques

One team created a Multilayer Perceptron (MLP) to classify the air quality into five different levels and then used K-Nearest Neighbours (K-NN) to predict to which outdoor activity location each sensor belongs. 

Another team did an EDA to find initial patterns in the data. Afterward, the stationarity and autocorrelation of one of the stations were analyzed by applying the Dickey-Fuller method. Finally, the future values were predicted by applying ARIMA.

Main Insights from Data

With simply an initial analysis, it was already possible to find where in the city the air quality is not as good - for example, Guincho had several indicators above the city's median in terms of Nitrogen Dioxide, PM10, and PM2.5. Another analysis showed how much time each parish spends under different levels of pollutants.


The teams that participated in this challenge presented a tool that could be used in the future to show locations with outdoor activities and the risk of air pollution (see Figure 1). Predicting the pollution levels in the future could also help regulate traffic and reduce the air quality health impact or to create targeted marketing campaigns to raise awareness among the population.

Figure 1 - An example of a dashboard showing the predictions of possible locations for outdoor activities.

Top 5 Solutions

1. Sidereus

2. BelManunel Soto

3. N/A

4. N/A

5. N/A

Open-source code

Other challenges