Institutional Repository of Vilnius University
Internetinių kontroliuojamų eksperimentų metrikų tarpusavio ryšių analizė /
Abstract
dc:descriptionIn online controlled experiments, the impact of product changes is most often evaluated based on the change in the primary success metric, while also monitoring that other important metrics do not deteriorate. However, a single metric does not always reveal the full context of the experiment results, since user behavior may be reflected in changes across several interrelated metrics. Therefore, this thesis examines how to select a compact group of metrics that is statistically related to the primary success metric and helps interpret experiment results more clearly. In this thesis, Lasso regression is applied for metric selection, as it enables simultaneous modeling of the primary success metric and selection of a smaller set of related variables. To assess the reliability of the selected metrics, the bootstrap method is additionally applied, and the results are validated using data from a second similar experiment. This research design makes it possible to evaluate not only the predictive performance of the model, but also the stability of the selected metrics across different experiments within the same product area. The results showed that Lasso regression can be used as a tool for selecting a compact group of metrics related to the primary success metric. The selected group of metrics remained stable when evaluated using different methods and validated on another similar experiment.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Girčys, Edvinas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD308334454&prefLang=en_US