PURBASARI, Intan Yuniar and Bayuseno, A.P. and Isnanto, R. Rizal and Winarni, Tri Indah (2026) PENGEMBANGAN TEKNIK EXPLAINABLE ARTIFICIAL INTELLIGENCE (XAI) BERBASIS ANCHOR UNTUK PENJELASAN PREDIKSI HASIL FUNGSIONAL SETELAH OPERASI PENGGANTIAN SENDI PANGGUL. Doctoral thesis, UNIVERSITAS DIPONEGORO.

	Text cover awal.pdf Download (200kB)
	Text 1_COVER.pdf Restricted to Repository staff only Download (868kB)
	Text 2_BAB_1.pdf Download (434kB)
	Text 3_BAB_2.pdf Download (1MB)
	Text 4_BAB_3.pdf Restricted to Repository staff only Download (616kB)
	Text 5_BAB_4.pdf Restricted to Repository staff only Download (1MB)
	Text 6_BAB_5.pdf Restricted to Repository staff only Download (338kB)
	Text 7_DAFTAR_PUSTAKA.pdf Download (330kB)
	Text 8_LAMPIRAN.pdf Restricted to Repository staff only Download (5MB)

Abstract

Salah satu tantangan utama dalam penggunaan sistem pendukung keputusan berbasis kecerdasan buatan (AI) atau machine learning (ML) adalah kemampuan untuk memberikan penjelasan (explainability) terhadap keputusan yang dihasilkan. Model ML umumnya bersifat black box, sehingga sulit dipahami dan divalidasi secara manusiawi. Bidang Explainable Artificial Intelligence (XAI) hadir untuk mengatasi hal ini dengan mengembangkan metode yang menjelaskan proses pengambilan keputusan model, terutama penting di domain medis yang menyangkut keselamatan pasien.
Penelitian ini mengembangkan teknik XAI berbasis Anchors, yaitu metode penjelasan berbentuk aturan if–then, dengan mengoptimasi hiperparameter utamanya (delta dan epsilon) menggunakan pendekatan Bayesian Optimization. Studi kasus diterapkan pada model prediksi hasil fungsional pasca-operasi Total Hip Arthroplasty (THA).
Hasil penelitian menunjukkan bahwa optimasi hiperparameter meningkatkan coverage—cakupan data yang dijelaskan oleh aturan anchor— sebesar 15% dengan tetap mempertahankan tingkat precision yang tinggi sebesar 97%. Pendekatan ini menunjukkan potensi penerapan metode XAI yang lebih adaptif, transparan, dan dapat dipercaya oleh klinisi. Ke depan, perluasan cakupan XAI terhadap data citra dan teks medis diharapkan dapat memperkuat validitas model dan mendukung pengambilan keputusan klinis yang lebih akurat dan berbasis data.
Kata kunci: Explainable Artificial Intelligence, penggantian sendi panggul, optimasi hiperparameter, algoritma Anchor

One of the main challenges in implementing artificial intelligence (AI) or machine learning (ML)–based decision support systems is ensuring explainability of the resulting models. ML models are inherently black boxes, making it difficult for humans to understand and validate their decision-making processes. Explainable Artificial Intelligence (XAI) has emerged to address this issue by developing methods and frameworks that enhance human understanding of AI models, which is especially critical in medical domains where patient safety is paramount.
This research develops an XAI technique based on the Anchors method—an interpretable, rule-based explanation approach—by optimizing its key hyperparameters (delta and epsilon) using Bayesian Optimization. The approach is applied as a case study to predict postoperative functional outcomes in Total Hip Arthroplasty (THA) patients.
Experimental results show that hyperparameter optimization improves coverage—the proportion of data for which an explanation applies— by 15% while maintaining good precision of 97%. The proposed method demonstrates the potential of adaptive, transparent, and clinically interpretable XAI systems to support medical decision-making. Future work should extend this framework to include image and textual medical data to strengthen predictive validity and further enhance trust in AI-assisted clinical applications.
Keywords: Explainable Artificial Intelligence, Total Hip Arthroplasty, hyperparameter optimization, Anchor algorithm

Actions (login required)

View Item