A Diagnostic Framework for Staged AI Adoption in Batik Motif Recognition: Integrating CNN Evidence and Implementation Readiness

Irwan Sembriring (Faculty of Information Technology, Satya Wacana Christian University)
Paminto Agung Christianto (Faculty of Information Technology, Institut Widya Pratama)
Eko Budi Susanto (Faculty of Information Technology, Satya Wacana Christian University Faculty of Information Technology, Institut Widya Pratama)
Suharyadi Suharyadi (Faculty of Information Technology, Satya Wacana Christian University)
Cheryl Louisa Loedwyca (Faculty of Information Technology, Satya Wacana Christian University)

Journal of Applied Data Sciences
Vol 7, No 3: September 2026

This study proposes a diagnostic dual-layer decision-support framework for staged artificial intelligence adoption in batik motif recognition. The objective is to examine whether technical evidence from convolutional neural network classification and perceived implementation-side readiness can be jointly interpreted to prevent premature deployment in cultural-heritage recognition. The contribution of the study is not a new classifier architecture, but an operational diagnostic logic that treats model performance, class-level instability, readiness perception, governance, security, and feedback mechanisms as complementary but non-substitutable evidence. Methodologically, the technical layer evaluated three transfer-learning baselines, namely VGGNet-16, ResNet50, and MobileNetV2, using 983 batik images across 20 motif classes. The implementation layer assessed perceived readiness among 173 information technology practitioners using the Technology-Organization-Environment-Human plus Feedback dimensions. The integration layer then mapped technical-readiness evidence and readiness perception into explicit staged-adoption decisions rather than averaging them as interchangeable indicators.  The analysis used performance summaries, readiness profiles, decision matrices, security checklists, learning curves, and confusion-matrix diagnostics to connect empirical observations with staged adoption recommendations. The best-performing baseline was ResNet50, with 45% accuracy and a macro F1-score of 0.40, showing low technical readiness and substantial motif-specific instability. In contrast, the readiness survey indicated high perceived implementation-side readiness, with an average agreement score of 78.7%. This mismatch reveals a readiness asymmetry: implementation support may exist even when the recognition model remains technically immature. The findings imply that batik-recognition systems should prioritize dataset expansion, expert label validation, model refinement, moderated feedback, security governance, and controlled pilot testing before operational deployment. The framework provides a transparent basis for risk-aware, staged adoption decisions in artificial-intelligence-assisted cultural heritage preservation.