A prominent approach in remote sensing poverty prediction uses deep learning satellite imagery with daytime and nighttime light (NTL). While results yield acceptable average predictions, certain areas consistently perform poorly. For high-stakes aid allocation decisions, however, achieving near-perfect predictions across all regions is crucial. To address this, we apply explainability techniques to evaluate vision models, enhance predictions with multimodal architectures, and investigate challenging cases. Our diagnostic analysis shows that fine-tuning with NTL tends to fail in areas where NTL poorly reflects wealth. Integrated Gradients and Guided Grad-CAM techniques also reveal that VGG, ResNet, and ViT architectures tend to focus on brightly lit features, such as roads and buildings. To complement vision models, a multimodal model integrates non-visual features. Metrics such as X (formerly known as Twitter) activity, distance from residential roads, and internet speed are especially predictive. Among 10 tested architectures, the Stack Ensemble Late Stage model, combining pre-trained weights, multimodal data, and NTL transfer learning, achieves the best performance with high accuracy and low error variance. Qualitative analysis of difficult-to-predict regions highlights the challenges of modeling wealth due to inherent aleatoric uncertainty. Two core issues, spatial inequality in small areas and the presence of rare features, underscore the need for principled cluster segmentation.

Smartphone overdependence among South Korean adolescents, affecting nearly 40%, poses a growing public health concern, with usage patterns varying by regional context. Leveraging conceptually informed AI/ML models, this study (1) develops a high-performing low-risk screening tool to monitor disease burden, (2) leverages AI/ML to explore psychologically meaningful constructs, and (3) provides place-based policy implication profiles to inform public health policy. This study uses data from 1873 adolescents in the 2023 Smartphone Overdependence Survey by the National Information Society Agency (NISA) in South Korea. Across the sample, the adolescents were about 14 years old (SD = 2.4) and equally distributed by sex (48.1% male). We then conceptually selected 131 features across two domains and 10 identified constructs. A nested modeling approach identified a low-risk screening tool using 59 features that achieved strong predictive accuracy (AUC = 81.5%), with Smartphone Use Case features contributing approximately 20% to performance. Construct-specific models confirmed the importance of Smartphone Use Cases, Perceived Digital Competence and Risk, and Consequences and Dependence (AUC range: 80.6–89.1%) and uncovered cognitive patterns warranting further study. Place-stratified analysis revealed substantial regional variation in model performance (AUC range: 71.4–91.1%) and distinct local feature importance. Overall, this study demonstrated the value of integrating conceptual frameworks with AI/ML to detect adolescent smartphone overdependence, offering novel approaches to monitoring disease burden, advancing construct-level insights, and providing targeted place-based public health policy recommendations within the South Korean context.

This is a corrigendum for the article “The Last Mile in Remote Sensing Poverty Prediction” published in ACM J. Comput. Sustain. Soc. 3, 3, Article 16 (June 2025), 54 pages.

This is an erratum for the article “The Last Mile in Remote Sensing Poverty Prediction” published in ACM J. Comput. Sustain. Soc. 3, 3, Article 16 (June 2025), 54 pages.