Master of Science – Biomedical Engineering, Thesis
Imputation of Missing Tabular Data Using Implicit Neural Representations (INRs) for Predictive Modelling.
Master of Science – Biomedical Engineering, Thesis Proposal
Imputation of Missing Tabular Data Using Implicit Neural Representations (INRs) for Predictive Modelling.
Real-world medical datasets frequently contain missing values due
to irregular patient follow-ups, incomplete records, or data entry
errors.
Accurate imputation of missing values is critical for creating reliable
predictions for various tasks. Traditional imputation methods like
mean or median substitution can introduce biases, while more
advanced techniques like KNN imputation or regression may not
fully capture the underlying patterns in the downstream task.
The aim of this thesis is to investigate the use of Implicit Neural
Representations (INRs) for imputing missing data in medical
records and to evaluate the impact of these imputations on
predictive models for tabular datasets. INRs, which represent data
as continuous functions rather than discrete values, offer a novel
approach to imputing missing values by learning smooth and dataaware representations.
The project will compare INR-based imputation to traditional
methods (e.g., mean imputation, KNN, or iterative imputation) and
assess the downstream effect of various tasks.
Nature of the Thesis
Programming: 80%, Documentation: 20%
Specific Requirements
• Experience in machine learning
• Good programming skills (especially Python)
Contact: anas.taha@unibas.ch