OPTIMIZATION OF DAILY FOOD COMBINATIONS FOR DIABETICS USING THE ARTIFICIAL IMMUNE SYSTEM ALGORITHM METHOD

Aditiya, Fajar Chandra (2026) OPTIMIZATION OF DAILY FOOD COMBINATIONS FOR DIABETICS USING THE ARTIFICIAL IMMUNE SYSTEM ALGORITHM METHOD. Undergraduate thesis, Universitas Pembangunan Nasional Veteran Jawa Timur.

	Text (cover) cover v4.pdf Download (721kB)
	Text (bab 1) 20081010090-BAB 1_organized.pdf Download (204kB)
	Text (bab 2) 20081010090-BAB 2.pdf Restricted to Repository staff only until 4 September 2028. Download (310kB)
	Text (bab 3) 20081010090-BAB 3.pdf Restricted to Repository staff only until 4 September 2028. Download (443kB)
	Text (bab 4) 20081010090-BAB 4.pdf Restricted to Repository staff only until 4 September 2028. Download (4MB)
	Text (bab 5) 20081010090-BAB 5.pdf Download (173kB)
	Text (daftar pustaka) 20081010090-DAFTAR PUSTAKA.pdf Download (176kB)
	Text (lampiran) 20081010090-LAMPIRAN.pdf Download (4MB)

Abstract

Food waste at the household level remains a critical issue, largely driven by the failure to utilize ingredients before they expire. This study develops an image-based recipe recommendation system that leverages the Artificial Immune System Algorithm (AISA) to detect ingredients nearing their expiration date. Inspired by the biological immune mechanism, AISA employs clonal selection and somatic hypermutation processes to classify visual features of 12 common food ingredient categories including eggs, chicken, tempeh, tofu, broccoli, carrots, and others stored in a refrigerator. A dataset of 6,000 images was collected from public sources and expanded through augmentation, with each image represented as a 512-dimensional feature vector combining HSV color histograms and Local Binary Pattern (LBP) texture descriptors. The system achieved a validation accuracy of 93.5% and a training accuracy of 94.72% after 100 evolutionary generations, outperforming CNN from scratch (82.3%), ResNet-50 (91.2%), and EfficientNet-B0 (92.1%), while also demonstrating superior training efficiency at 3.2 hours compared to all baselines. Real-world testing under varied lighting conditions yielded an average accuracy of 88.4%, confirming adequate generalization capability for practical deployment. Explainability analysis through affinity-based feature contribution maps revealed that classification errors between visually similar classes particularly tempeh and tofu stem from near-identical HSV color distributions and LBP texture patterns, providing a clear direction for future refinement. The system is integrated into a mobile application capable of detecting ingredients from images and recommending recipes that prioritize the use of near-expiry items, offering a practical digital solution to reduce household food waste.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Parlika, Rizky	0718058401	rizkyparlika.if@upnjatim.ac.id
Thesis advisor	Sihananto, Andreas Nugroho	0012049005	andreas.nugroho.jarkom@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA76.6 Computer Programming
R Medicine > RZ Other systems of medicine

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Fajar Chandra Aditiya

Date Deposited:

07 Sep 2026 02:36

Last Modified:

07 Sep 2026 04:43

URI:

https://repository.upnjatim.ac.id/id/eprint/59990