KLASIFIKASI TINGKAT PEMROSESAN MAKANAN BERBASIS TEKS KOMPOSISI MENGGUNAKAN STRATEGI WEIGHTED ENSEMBLE LARGE LANGUAGE MODELS

Khairul Hudha Nasution; Indra Budi

doi:10.24252/instek.v11i1.64305

Authors

Khairul Hudha Nasution Universitas Indonesia
Indra Budi Universitas Indonesia

DOI:

https://doi.org/10.24252/instek.v11i1.64305

Keywords:

ensemble learning, large language models, ultra-processed foods

Abstract

Klasifikasi tingkat pemrosesan makanan merupakan langkah krusial dalam mitigasi risiko kesehatan global akibat konsumsi makanan ultra-proses. Meskipun label komposisi tersedia pada kemasan, penulisan yang kecil, tidak terstruktur, typo serta terminologi kimia yang kompleks memiliki kecenderungan menyulitkan penilaian manual oleh konsumen. Pemanfaatan Large Language Models (LLM) menawarkan potensi efisiensi deteksi otomatis, namun mengandalkan satu arsitektur model tunggal memiliki risiko tinggi akibat variabilitas performa. Penelitian ini bertujuan untuk mengevaluasi efektivitas strategi Weighted Ensemble Learning dibandingkan model tunggal dalam memprediksi skor pemrosesan makanan (FPro) pada dataset GroceryDB. Eksperimen dilakukan menggunakan lima arsitektur LLM dengan skala parameter kecil hingga sedang (Gemma-3-4B, Llama-3.2-3B, Qwen3-4B, R1-Distill-1.5B, dan Phi-2) melalui pendekatan Weighted Voting berbasis kinerja historis. Hasil evaluasi menunjukkan adanya disparitas ekstrem pada kinerja model tunggal, di mana model berkapasitas rendah (Phi-2) mengalami kegagalan penalaran dengan F1-score hanya 10%, sementara model dengan kemampuan instruksi tinggi (Gemma) mencapai 68%. Penerapan strategi Ensemble berhasil memitigasi kelemahan model individual melalui mekanisme koreksi silang, meningkatkan akurasi F1-score menjadi 70% dan menghasilkan prediksi yang lebih stabil serta robust dibandingkan jika hanya mengandalkan satu model terbaik sekalipun.

Downloads

Download data is not yet available.

References

[1] B. Srour et al., "Ultra-processed food intake and risk of cardiovascular disease: Prospective cohort study (NutriNet-Santé)," The BMJ, vol. 365, p. l1451, 2019.

[2] C. A. Monteiro et al., "Ultra-processed foods: what they are and how to identify them," Public Health Nutrition, vol. 22, no. 5, pp. 936-941, 2019.

[3] Y. Guo, A. Ovadje, M. A. Al-Garadi, and A. Sarker, "Evaluating large language models for health-related text classification tasks with public social media data," Journal of the American Medical Informatics Association, vol. 31, no. 10, pp. 2181–2189, 2024.

[4] A. Kamen and Y. Kamen, "Majority rules: LLM ensemble is a winning approach for content categorization," arXiv preprint arXiv:2511.15714, 2025.

[5] B. Ravandi, G. Ispirova, M. Sebek, P. Mehler, A.-L. Barabási, and G. Menichetti, "Prevalence of processed foods in major US grocery stores," Nature Food, vol. 5, pp. 1-9, 2024.

[6] A. Vaswani et al., "Attention is all you need," in Advances in Neural Information Processing Systems, vol. 30, 2017, pp. 5998-6008.

[7] W. X. Zhao et al., "A survey of large language models," arXiv preprint arXiv:2303.18223, 2023.

[8] Gemma Team, "Gemma: Open models based on Gemini research and technology," Google DeepMind, London, UK, Tech. Rep., 2024.

[9] A. Dubey et al., "The Llama 3 Herd of Models," arXiv preprint arXiv:2407.21783, 2024.

[10] DeepSeek-AI, "DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning," arXiv preprint arXiv:2501.12948, 2025.

[11] J. Bai et al., "Qwen Technical Report," arXiv preprint arXiv:2309.16609, 2023.

[12] M. Javaheripi, S. Bubeck, et al., "Phi-2: The surprising power of small language models," Microsoft Research Blog, Dec. 2023. [Online]. Available: https://www.microsoft.com/en-us/research/blog/phi-2-the-surprising-power-of-small-language-models/

[13] J. Wei et al., "Chain-of-thought prompting elicits reasoning in large language models," in Advances in Neural Information Processing Systems, vol. 35, 2022, pp. 24824-24837.

[14] Z. Ji et al., "Survey of hallucination in natural language generation," ACM Computing Surveys, vol. 55, no. 12, pp. 1-38, 2023.

[15] X. Wang et al., "Self-consistency improves chain of thought reasoning in language models," in International Conference on Learning Representations (ICLR), 2023.

[16] D. Jurafsky and J. H. Martin, Speech and Language Processing, 3rd ed. Upper Saddle River, NJ, USA: Pearson, 2021.