Closing the design phase of a Bogota air-quality project. The team will estimate missing sensor values, interpolate PM2.5 between stations, and build a web/mobile app to show real-time and historical data. They chose mean absolute error to evaluate the model and flagged risks like surveillance and criminalization before moving to implementation.
Congratulations on completing the design phase for the Bogota Air Quality project. In this phase of the project, you're prototyping your solution, considering how to deal with any issues in the data regarding privacy or personal information, which in this case should be minimal, and you're designing the user experience. Nice work. At this point, you should be able to answer the following questions before moving on to implementation. First, how you address issues with imbalances, biases, privacy or other concerns with your data. What kind of model will you implement and how will you measure its performance? How will your design address the problem you set out to solve? And how will the end user interact with your system? As mentioned before, you always need to do a thorough analysis of data regarding data privacy, security bias or other concerns. For this project, you work with publicly available data from scientific instruments, so there aren't any major concerns regarding protecting personal information. Keep in mind, however, that just the fact that a data set is publicly available, that doesn't mean that it can do no harm. The application that you are working on is intended to be part of a broader public health initiative regarding air quality in the city of Bogota. So you need to be sure that at the very least, your product does no harm by providing wrong or confusing information. And I think you can consider some other potential misuses of this information that you could discuss with stakeholders. For example, if you were able to come up with much more accurate estimates of sudden bursts of pollution, this could be associated with illegal manufacturing, or an illegal burn off, or some other criminal activity that could attract surveillance use cases. Not just surveillance for pollutants, but surveillance for crime. And if this information is then used to prosecute people for crimes or identify criminals, then the community might understandably lose trust in these sensors and there will be resistance towards trust in the sensor data or even having future sensors in installed. And so, for any use case, even if you know it doesn't have personal information to begin with, you should spend time with the stakeholders thinking about potential risks and what are the potential downsides of providing a richer experience with machine learning through the lab. Here you came up with a model designed for both estimated missing values in sensors and estimating the PM2.5 between the sensors. And you use the mean absolute error to establish your model's performance against the baseline. With this design, we aim to address the problem you set out to work on by providing the city of Bogota with an application that allows citizens to get real time estimates of air quality throughout the city. And those end users will interact with your system via a web or mobile application. This will then allow them to see current estimates as well historical data from individual census data. Now, of course, the actual design that you've been working on in the labs is still very much a prototype. But over the course of these labs, you've been able to demonstrate a system that can estimate missing sensor values, estimate pollution levels in between sensors, and provide a snapshot of historical data if you click on a sensor station location. And with that, you've completed the design phase for this project. Now it's time to put it all together in the implementation phase. So please join me for the next video to walk through the lab section on implementing your product.
