Sistem Pakar Diagnosa Penyakit Tembakau Berbasis Web dengan Pendekatan Graph-Based Reasoning menggunakan Knowledge Graph dan Graph Convolutional Network

Abstract

A tobacco disease diagnosis expert system was implemented using a Knowledge Graph–Graph Convolutional Network (KG-GCN). Disease–symptom relationships were encoded in a Knowledge Graph and processed through a GCN with attention pooling to generate top-k diagnostic predictions. The approach achieved strong results with Hits@5 = 1.00, MRR = 0.93, and AUPR = 0.93. System-level evaluation confirmed correct end-to-end functionality, while expert validation across 20 scenarios showed full agreement with specialist diagnoses. These results indicate that KG-GCN provides an effective framework for early tobacco disease diagnosis.