• Harsha B K

Purpose: This paper develops a research framework for evaluating AI-based civic-issue classification beyond headline accuracy. Civic reporting systems can help public agencies identify potholes, waste accumulation, damaged streetlights, drainage failures, and other local infrastructure problems from citizen-submitted images. However, a model that performs well on a test set may fail when images are blurred, poorly illuminated, compressed, partially occluded, or captured in unfamiliar locations. A further concern is whether explanations offered by explainable artificial intelligence (XAI) methods faithfully represent the evidence used by a model. Design/methodology/approach: The paper proposes a reproducible comparative evaluation of image-classification models using a documented public dataset or a carefully governed civic-issue dataset. A conventional convolutional neural network is compared with a transfer-learning baseline and, where computational resources permit, a modern vision model. Evaluation includes macro-F1, per-class precision and recall, confusion matrices, calibration, inference latency, and controlled corruption tests. Explanation methods such as Grad-CAM are assessed using deletion/insertion or perturbation-based faithfulness checks and stability analysis rather than visual inspection alone. Expected contribution: The framework treats predictive performance, robustness, explanation quality, and operational efficiency as related but distinct dimensions. It proposes a reporting protocol that makes model limitations visible to researchers and public-sector users. No new experimental findings are claimed in this manuscript; numerical results must be produced after implementation and validation. The paper offers a feasible route from a student software project to a defensible empirical study, with relevance to responsible AI, computer vision, and digital public services.

Explainable Artificial Intelligence, Computer Vision, Civic Issue Classification, Model Robustness, Grad-CAM, Responsible AI, Public Services, Image Classification

"Beyond Accuracy: Evaluating Explainability and Robustness in AI-Based Civic Issue Classification", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.13, Issue 10, page no.c14-c21, October-2026, Available :http://www.jetir.org/papers/JETIR2610203.pdf

"Beyond Accuracy: Evaluating Explainability and Robustness in AI-Based Civic Issue Classification", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.13, Issue 10, page no. ppc14-c21, October-2026, Available at : http://www.jetir.org/papers/JETIR2610203.pdf