IMPUTASI DATA DERET WAKTU MULTIVARIAT MENGGUNAKAN METODE SKDL (STING KERNEL DEEP LEVEL) WITH EXPLAINABLE

KURNIAWANSYAH, ARIUS SATONI and Nurmaini, Siti and Erwin, Erwin (2025) IMPUTASI DATA DERET WAKTU MULTIVARIAT MENGGUNAKAN METODE SKDL (STING KERNEL DEEP LEVEL) WITH EXPLAINABLE. Doctoral thesis, Sriwijaya University.

	Image RAMA_21001_03013682126021_cover.jpg - Cover Image Available under License Creative Commons Public Domain Dedication. Download (179kB)
	Text RAMA_21001_03013682126021.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (5MB) | Request a copy
	Text RAMA_21001_03013682126021_TURNITIN.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (10MB) | Request a copy
	Text RAMA_21001_03013682126021_0002085908_0029017101_01_front_ref.pdf - Accepted Version Available under License Creative Commons Public Domain Dedication. Download (993kB)
	Text RAMA_21001_03013682126021_0002085908_0029017101_02.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (662kB) | Request a copy
	Text RAMA_21001_03013682126021_0002085908_0029017101_03.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (1MB) | Request a copy
	Text RAMA_21001_03013682126021_0002085908_0029017101_04.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (1MB) | Request a copy
	Text RAMA_21001_03013682126021_0002085908_0029017101_05.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (15kB) | Request a copy
	Text RAMA_21001_03013682126021_0002085908_0029017101_06_ref.pdf - Bibliography Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (298kB) | Request a copy
	Text RAMA_21001_03013682126021_0002085908_0029017101_07_lamp.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (1MB) | Request a copy

Abstract

This study aims to develop a new method to address the problem of missing values in multivariate time series data, particularly in vital sign data of ICU patients derived from the MIMIC-IV dataset. Missing data in the medical context can hinder analysis, prediction, and even clinical decision-making. Therefore, an accurate and explainable imputation approach is crucial. This research introduces a combined approach involving two methods: the STING method (Self-Attention-based Time Series Imputation using GAN) and the Enhanced Kernel method. The Enhanced STING method is designed to generate precise numerical imputations using attention mechanisms and a GAN architecture, while the Kernel method focuses on achieving better imputations in the medical domain. The combination of these two methods is referred to as SKDL (STING Kernel Deep Level with Explainable), an imputation method capable of producing dual outputs (numerical and categorical) and equipped with Explainable AI components. The model is evaluated using MAE, MSE, RMSE, R-Squared, Sensitivity, Specificity, and F1-Score as performance metrics. The results show that the Enhanced STING method achieved a MAE of 0.090 and RMSE of 0.0109, outperforming existing methods such as KNN, MICE, and Mean Imputation. Meanwhile, the Enhanced Kernel method achieved an F1-Score of 0.825. The SKDL method achieved MAE of 0.0870, MSE of 0.0175, RMSE of 0.0040, and R-Squared of 0.3367, which shows the best accuracy results compared to the Advanced Stage STING and Advanced Stage Kernel Methods. These findings indicate that the proposed approach not only produces more accurate imputations but also provides clinically interpretable outputs in the form of categorical data representing patient condition status. This research provides a significant contribution to the field of medical data science, particularly in the development of AI-based decision support systems for intensive care units. Keywords: Imputation, Time Series, Multivariat, Data, STING, Kernel, Deep Level, Explainable

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Imputation, Time Series, Multivariat, Data, STING, Kernel, Deep Level, Explainable
Subjects:	T Technology > T Technology (General) > T1-995 Technology (General) T Technology > T Technology (General) > T1-995 Technology (General) > T18 Modern T Technology > T Technology (General) > T1-995 Technology (General) > T180.A1 General works
Divisions:	03-Faculty of Engineering > 21001-Engineering Science (S3)
Depositing User:	Arius Satoni Kurniawansyah
Date Deposited:	23 Sep 2025 03:00
Last Modified:	23 Sep 2025 03:00
URI:	http://repository.unsri.ac.id/id/eprint/184572

Actions (login required)

View Item