Date of Submission

Fall 2023

Supervisor

Dr. Tariq Mahmood, Professor, Department of Computer Science, School of Mathematics and Computer Science (SMCS), Institute of Business Administration (IBA), Karachi

Committee Member 1

Dr. Rizwan Ahmed Khan, Professor and Dean, Department of Computer Science, Salim Habib University, Karachi, Pakistan.

Committee Member 2

Dr. Shahid Hussain, Associate Professor and Chairperson, Department of Computer Science, School of Mathematics and Computer Science (SMCS), Institute of Business Administration (IBA), Karachi

Degree

Master of Science in Data Science

Department

Department of Computer Science

Faculty/ School

School of Mathematics and Computer Science (SMCS)

Keywords

Mortality Prediction, Deep Learning, Ensemble, Machine Learning, Stacking

Abstract

PICU mortality is a critical issue, and its prediction is essential. In this thesis, our primary objective was to construct a mortality prediction framework for patients in the PICU using deep learning ensemble methods, following extensive experimentation on high-frequency vital signs and laboratory test data. The proposed framework, named PEDICTOR, is a unique and novel design specifically tailored to PICU patients in the tertiary care hospital of Aga Khan University (AKU) in Pakistan. We developed a novel feature selection approach that incorporates expert domain knowledge and sets it apart from other methods. Another distinctive feature of PEDICTOR is that it is trained on specific age-based groups of pediatric patients and provides predictions based on the age group in which the patient lies. To address the problem of class imbalance in the data, we incorporated a Cluster-Based Synthetic Minority Oversampling (SMOTE) technique. After rigorously testing various Machine Learning (ML) and Deep Learning (DL) models, we experimented with various stacker combinations involving ML and DL models such that our PEDICTOR framework contains the optimal stacker model. The PEDICTOR framework provided an overall mortality prediction with an F1-score in the range of (0.441–0.740) for expired patients and an AUCROC in the range of (0.846–0.932) for PICU patients, thereby enabling doctors to save lives and hospitals to manage their resources efficiently.

Document Type

Restricted Access

Submission Type

Thesis

Available for download on Thursday, June 10, 2027

Share

COinS