Generering och visualisering av tidsserieprediktioner
Typ
Examensarbete pÄ grundnivÄ
Program
Publicerad
2019
Författare
Wester, Anton
Nilsson, Oscar
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
This project was established together with an IoT-company that manufactures and sells products for home automation and saves a lot of data from their installed products which is not used in company-related decisions. The aim of the project is to investigate whether there is a developed API or a method for creating forecasts on the company's historical data in order to be able to make more realistic business decisions. The technical parts of the work have been carried out in the company's premises, where there has been access to computers and consultation of experienced developers, while the "non-technical" parts have been performed at Chalmers.
An API for the creation of time series forecasts was discovered after weeks of investigation. The API is created by BigML and provides functionality to load data, process the data and produce forecasts. After the group decided to use BigML, they implemented its functionality in to the company's existing analysis tool and visualized the historical data along with the forecasted data in a graph.
The group encountered a number of problems during the project, but are generally satisfied with the result. A clear representation of the historical data and the forecast data was successfully implemented in the company's code base, but there is uncertainty about the reliability of the forecasted values.