Användning av maskininlärning för identifiering av källare med hjälp av byggnadsdata

Sammanfattning: The study investigates whether it is possible to use machine learning (ML) and existing building data to identify whether a building has a basement or not. The study discusses various steps needed for the development of the machine learning models such as data collection, data preparation and implementation of the models using appropriate frameworks as well as analysis and evaluation of results.  Design Science has been used as the methodology to conduct the study in a structured manner where the process follows the steps of the methodology. The study describes identification of relevant data sources, data collection, data preparation, data annotation and training, testing, and evaluation of the models.  Three ML models are implemented with the application of convolutional neural network (CNN), artificial neural network (ANN) and support vector machine (SVM). All models have been tested and adjusted to achieve the best possible accuracy. 20 new addresses that were not included in the training of the models were then generated. The purpose of these addresses was to test the three models with the same data, to compare the ability of the different models to identify basements in buildings. It was found that ANN and SVM both generated relatively high correct classifications while CNN generated low correct classifications. Evaluation of the models shows that all ML models require more data to achieve a more reliable and accurate classification model for building identification.