Studi Komparatif Metode Counterfactual dan Example-Based Dalam Meningkatkan Explainability Pada Antarmuka Dashboard Monitoring Karung Pakan

Ananda Kusuma Putri, Divi Galih Prasetyo Putri, S.Kom., M.Kom., Ph.D.

2026 | Tugas Akhir | D4 Teknologi Perangkat Lunak

Perkembangan Artificial Intelligence (AI) telah mendorong pemanfaatan sistem prediksi untuk mendukung pengambilan keputusan, termasuk pada pengelolaan pakan ayam broiler. Namun, sifat black-box pada model machine learning dapat menurunkan pemahaman dan kepercayaan pengguna nonteknis terhadap hasil prediksi. Meskipun Explainable Artificial Intelligence (XAI) telah banyak diteliti, sebagian besar berfokus pada domain kesehatan dan finansial. Penelitian ini bertujuan membandingkan efektivitas metode counterfactual dan example-based pada dua variasi antarmuka XAI dibandingkan dengan antarmuka baseline tanpa XAI dalam meningkatkan usability, understandability, dan trustworthiness pada domain smart farm. Evaluasi dilakukan melalui eksperimen within-subject terhadap 21 partisipan yang menyelesaikan serangkaian tugas pada ketiga antarmuka. Hasil penelitian menunjukkan bahwa metode counterfactual paling efektif meningkatkan usability dan trustworthiness. Metode example-based memperoleh skor understandability tertinggi pada pengguna dengan peran administrasi. Temuan ini memberikan evidensi empiris efektivitas metode penjelasan XAI pada dashboard berbasis AI di domain smart farming.

The development of Artificial Intelligence (AI) has driven the adoption of predictive systems to support decision-making, including in broiler chicken feed management. However, the black-box nature of machine learning models can reduce non-technical users' understanding of and trust in prediction results. Although Explainable Artificial Intelligence (XAI) has been extensively studied, most research has focused on the healthcare and financial domains. This study aims to compare the effectiveness of counterfactual and example-based explanation methods across two XAI interface variations against a baseline interface without XAI in improving usability, understandability, and trustworthiness in the smart farming domain. The evaluation was conducted through a within-subject experiment involving 21 participants who completed a series of tasks using all three interfaces. The results show that the counterfactual method was the most effective in improving usability and trustworthiness, while the example-based method achieved the highest understandability scores among users with administrative roles. These findings provide empirical evidence regarding the effectiveness of XAI explanation methods in AI-based dashboards within the smart farming domain.

Kata Kunci : Counterfactual explanation, example-based explanation, Explainable Artificial Intelligence (XAI), smart farm, UI/UX