Citation

Shaaban, Mohammed Najib Ahmed (2025) Spatial analysis and risk modelling of disease outbreak using satellite imagery. Masters thesis, Multimedia University.

Full text not available from this repository.

Abstract

The rapid global spread of COVID-19, declared a pandemic by the World Health Organization on March 11, 2020, spurred extensive efforts by governments and stakeholders to manage the crisis and pave the way for recovery. In this context, satellite technology has proven invaluable in epidemiological studies, enabling the quantification of epidemic-related factors. While the demand for pandemic-related satellite studies has escalated, most research has focused on evaluating the effects of lockdowns on air quality, economic activity, and traffic patterns. However, there has been limited exploration of risk-based predictive modeling that could offer governments long-term, sustainable solutions. In response to this gap, we developed a land-cover-based risk assessment framework utilizing satellite imagery to estimate and predict COVID-19 risk levels at localized geographical scales. Daily COVID-19 case data were sourced from the Ministry of Health, Malaysia, while satellite imagery was acquired via the Google Static Maps API. A critical challenge identified was the lack of publicly available tropical land cover datasets suitable for deep learning. To address this, we created and annotated "DeepEarthMY," a dataset specifically designed for tropical environments, and trained a semantic segmentation model to extract land cover features. This model facilitated the exploration of the correlation between various land cover elements and COVID-19 prevalence. To fulfill the primary objective of developing a precise risk assessment model, we proposed a novel two-stream end-to-end deep learning framework that effectively integrates multiple data sources. The first stream extracts spatial and contextual features directly from raw satellite imagery using a Convolutional Neural Network (CNN). The second stream processes numerical land-cover statistics derived from the segmentation model, specifically the proportional composition of key environmental features such as buildings, forests, and water bodies. The outputs from both streams are fused through a joint learning layer, enabling the model to capture complex spatial–statistical relationships that single-input models often overlook. We evaluated the effectiveness of this proposed model by comparing it against three established machine learning classifiers: Support Vector Machine (SVM), Random Forest (RF), and Decision Tree (DT). Our model demonstrated superior performance, achieving an accuracy of 85.2%. These findings illustrate the significant role of land cover factors in influencing the spread of COVID-19 at the local level, underscoring the potential of satellite imagery in epidemiological risk assessment.

Item Type:	Thesis (Masters)
Additional Information:	Call No.: Q325.5 .S53 2025
Uncontrolled Keywords:	Machine learning—Industrial applications
Subjects:	Q Science > Q Science (General) > Q300-390 Cybernetics
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	05 Oct 2026 06:59
Last Modified:	05 Oct 2026 06:59
URII:	http://shdl.mmu.edu.my/id/eprint/16884

Downloads

Edit (login required)