Capturing the Base Station by Feature Engineering

Abstract

A feature vector is a compact representation of an object or system, that contains the most important or informative aspects of the entity. A feature vector is often used in the context of machine learning due to the effectiveness and noise reduction, but can also be used for data exploration. In this thesis we want to investigate if a complex system such as a base station can be represented with a feature vector in an useful manner. To create the feature vector, we ask subject matter experts for the most relevant attributes of the base station. The feature vector is evaluated by partitioning all the base stations, either with an unsupervised clustering algorithm, or by an interesting attribute of the base station. The partitions are then visualized and presented to experts, who determined if the partitions brings forth interesting patterns that can be useful, or if the partitions are useful in themselves. The result is a feature vector containing 494 features, based of 22 attributes that are recommended by subject matter experts. The feature vector brings forth interesting and useful patterns, and we can thus conclude that a feature vector can be used to represent a complex system such as a base station in an useful manner.