The main goal was to get educated in machine learning and determine a suitable development platform and toolchain.
Prediction of energy consumption in real estate was chosen as the use case, as such data is readily available via APIs and for multiple years back. There were other options like theft and fraud detection that were much harder to get enough data for.
Python was chosen as the platform and language for machine learning as the goal was to make a web application. Initially Anaconda and Jupyter were used as the development environment, but were eventually replaced by Visual Studio Code and standalone/independent application development.
During the project several APIs for energy and weather data were investigated practically, including Metry, E.ON, OpenWeather, SMHI and Trafikverket. Abiro also had access to the API provided by Sensative’s Yggio platform.
To experiment with energy, weather data as well as different machine learning methods and models, an experimental site was set up at AI for Real Estate (AIRE). It was developed in PHP before realizing Python (or R) was the way to go for machine learning overall.
Several spin-off projects are considered for the future based on the outcomes. Offerings available so far are Abiro Meters and Abiro Energy Prophet, both using the same energy data. Abiro Weather and Abiro Stargazer are concrete uses of weather data.