Pembobotan Adaptif pada Data Multidimensi dengan Multi-Agent Reinforcement Learning untuk Rekomendasi Kinerja Staf Akademik

Tarigan, Rangga Satya (2026) Pembobotan Adaptif pada Data Multidimensi dengan Multi-Agent Reinforcement Learning untuk Rekomendasi Kinerja Staf Akademik. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025232019-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Evaluasi kinerja staf akademik mencakup aspek pengajaran, penelitian, dan pengabdian kepada masyarakat. Penilaian terhadap aspek tersebut masih menggunakan bobot yang seragam pada seluruh program studi, padahal kebutuhan dan pola kontribusi setiap rumpun ilmu berbeda. Kondisi ini dapat membuat prioritas peningkatan kinerja tidak tepat sasaran, dimana aspek yang perlu diperkuat pada satu program studi tidak selalu sama dengan program studi lainnya. Penelitian ini mengembangkan pembobotan adaptif dengan Multi-Agent Reinforcement Learning (MARL) untuk mempelajari penyesuaian bobot berdasarkan konteks rumpun ilmu dan pola data kinerja staf akademik. Data bersumber dari system internal Universitas Medan Area periode 2022-2024 dengan 1.495 data staf akademik. Indikator kinerja dirangkum ke dalam lima dimensi yaitu kualitas pengajaran, kuantitas pengajaran, penelitian, inovasi, dan pengabdian masyarakat. Metode MARL dengan algoritma Proximal Policy Optimization digunakan karena dapat mempelajari pola data dan menyesuaikan strategi bobot secara adaptif, sehingga kebijakan pembobotan menjadi lebih spesifik dan dapat mengarahkan prioritas pada aspek yang paling tertinggal. Hasil penelitian menunjukkan metode adaptif memberikan nilai lebih baik dibanding metode statis, dengan imbalan kumulatif 12,75 dan nilai Gini di atas 0,5, selain itu menghasilkan sebaran bobot yang berbeda antar program studi yang menyesuaikan dengan konteks rumpun ilmu. Temuan ini menunjukkan pembobotan adaptif berbasis MARL dapat menjadi dasar untuk mendukung keputusan peningkatan kinerja staf akademik.
===================================================================================================================================
Academic staff performance evaluations cover teaching, research, and community service. These aspects are assessed using uniform weighting across all study programs, despite the varying needs and contribution patterns of each discipline. This can lead to misdirected performance improvement priorities, as aspects that need to be strengthened in one study program are not always the same as those in another. This research developed adaptive weighting using Multi-Agent Reinforcement Learning (MARL) to explore weight adjustments based on the context of the discipline and the data patterns of academic staff performance. Data were sourced from the internal system of Medan Area University for the 2022-2024 period, comprising 1,495 academic staff. Performance indicators are summarized into five dimensions: teaching quality, teaching quantity, research, innovation, and community service. The MARL method, with its Proximal Policy Optimization algorithm, is used because it can learn from data patterns and adapt weighting strategies, resulting in more specific policy weighting and prioritizing the most lagging aspects. The results showed that the adaptive method provided better results than the statistical method, with a cumulative imbalance of 12.75 and a Gini value above 0.5. It also produced different weight distributions between study programs, adapting to the context of the scientific cluster. These findings suggest that MARL-based adaptive weighting can be a basis for supporting decisions to improve academic staff performance.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Pembobotan Adaptif, Multi-Agent Reinforcement Learning, Evaluasi kinerja staf akademik, Proximal Policy Optimization, Adaptive Weighting, Multi-Agent Reinforcement Learning, Academic Performance, Proximal Policy Optimization
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Rangga Satya Tarigan
Date Deposited:	29 Jan 2026 07:11
Last Modified:	29 Jan 2026 07:11
URI:	http://repository.its.ac.id/id/eprint/130941

Actions (login required)

View Item