Fischer, Richard (2023) Development and Evaluation of Causal Models With an Application to Orthopaedic Inquiries. Master's Thesis / Essay, Artificial Intelligence.
|
Text
RFischer_MasterThesis.pdf Download (636kB) | Preview |
|
Text
toestemming.pdf Restricted to Registered users only Download (125kB) |
Abstract
Several influential leaders in the field of Artificial Intelligence argue that something is missing in the current Deep Learning approach: causal understanding. In the field of causal reasoning, models that quantify the causal effect of some variable on another are developed. This thesis offers an overview of the field, a comparison of several estimators, including an extension to the X-Learner coined X-Learner++ and an ensemble consisting of three estimators. The comparison is done on the semi-synthetic IHDP dataset. Furthermore, a medical dataset to investigate the effect of prophylactic treatment on Venous Thromboembolism (VTE) incidence is analyzed and compared to a peerreviewed meta-analysis. The best-performing estimator in the IHDP dataset is the Augmented Inverse Probability Weighting (AIPW) estimator and the X-Learner++ with an error of -0.09. In the medical data set, the ensemble approach worked best when compared to the meta-analysis with an error of 0.51%. The thesis introduces the causal reasoning methodology to the realm of orthopaedics, and establishes trust by successfully emulating an existing meta-analysis. Furthermore, we establish that the estimators, and the developed error correction for the X-Learner, work well on the semi-synthetic dataset.
Item Type: | Thesis (Master's Thesis / Essay) |
---|---|
Supervisor name: | Jaeger, H. |
Degree programme: | Artificial Intelligence |
Thesis type: | Master's Thesis / Essay |
Language: | English |
Date Deposited: | 26 Jan 2023 09:39 |
Last Modified: | 26 Jan 2023 09:39 |
URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/29158 |
Actions (login required)
View Item |