An air quality monitoring project in Bogotá uses AI to estimate PM2.5 pollution. The model's error rate is 4-7 µg/m³, and the team emphasizes testing with actual users and engaging stakeholders before scaling to other regions like South Africa.
You're now at a point where you've explored, designed and implemented your air quality monitoring system. Nice work. At this point, before deploying your system out into the real world, there are a couple of questions that you'll need to address before moving forward. The two questions that you'd like to be able to answer in the affirmative at this stage are is your model performance acceptable? And are the end users able to successfully interact with your system? In this case, you found that your models for both estimating missing values as well as estimating pollution in between the sensors have associated errors on the order of 4 to 7 micrograms per meter cubed when estimating PM2.5 levels. Having an estimate with this sort of associated error is most likely better than having no measurement at all. And for the purposes of advising the public or whether the air is safe to breathe, we we can make the assumption in this course that this is a reasonable margin of error. Keep in mind that different error estimates, like mean squared error or accuracy measures like precision and recall might be more appropriate for different kinds of projects. And this will all depend on your particular project goals and the individual specifications and stakeholders. When it comes to how end users are interacting with your system, you would have to do some end user testing to see how things are going. Are they comfortable with the way that they're interacting with the software? Are they interacting with it in the way that you thought that they would, as well as talk to other stakeholders to learn whether your product is sufficient to serve their needs. Depending on what you find when you speak to the end users and other stakeholders, you might decide to work on an updated implementation, develop a new design for your system, or start over to explore another aspect of the problem of air pollution or or perhaps some other problem entirely. While working on estimating the air quality in Bogota, you went through all of the steps of our AI for Good framework. In the Explore phase, you looked at who the relevant stakeholders might be. In a project like this, we've included public health officials, policymakers, people managing the air quality sensors and the citizens. You formulated a problem statement and you explored the data to determine whether it looked like AI could in fact act add value to this project. In your exploration of the data, you uncovered patterns including correlations between different pollutants, as well as trends in pollutant levels that depended on things like the time of day, week and the sensor station location. And all of this gave you a strong indicator that AI may in fact be able to add value here. In the design phase, you investigated missing data more deeply and developed models to estimate missing sensor measurements. First a simple baseline model and then a neural network that performed much better. After using your models to estimate missing values across all the pollutants in the data set, you developed a method for estimating pollution levels in between the sensor stations. In the implementation phase, you put your models to work, generated maps that would be useful to the end user when it comes to knowing what pollutant levels are right now, as well as how pollution levels might have been changing in recent days or hours across the city. If you're feeling inspired to keep working on some of the challenges that you encountered in the air quality monitoring project, then I think that's great. You could think about things like downloading more data from the city of Bogota or even looking for some air quality data for the area where you live. At this point, I'd also encourage you to take a moment to think about what kind of projects you might be interested in and how might you approach those projects using the same framework that we've been working on throughout this course. So, for example, you could be thinking about, well, who would the key stakeholders be and how would I contact them? What kind of problem would you be addressing? Is there likely to be a data component to this problem? And with these questions in mind, you're well on your way to getting started. Before you wrap up this course, we have another project Spotlight for you. Tepewa Chawerewe is a technology strategist who has worked in air pollution in South Africa. So, like we said, you could take this project that you worked on for Bogota, Colombia, and then apply it to any other location. In the next video, you'll hear how Tapioa developed a solution in his home country of South Africa.
