Estimation of Fireproof Structure Class and Construction Year for Disaster Risk Assessment

TL;DR

The paper tackles disaster-risk profiling in Japan by estimating critical building attributes—construction year, building structure, and property type—from exterior facade images to derive the fireproof class (H/T/M) for insurance purposes. It introduces a multi-task learning framework with a ResNet-101 backbone that jointly predicts construction year (regression), building structure and property type (classification), and then maps these intermediate attributes to the fireproof class using a rule-based scheme aligned with GIROJ criteria. On the LIFULL_HOMES dataset, the approach achieves construction-year MAE of about $4.97$ years and RMSE of about $6.88$, structure accuracy around $92.75\%$, property-type accuracy around $83.21\%$, and fireproof-class accuracy near $89\%$, with notable robustness despite data imbalance and intermediate prediction errors. These results demonstrate scalable, interpretable image-based risk profiling suitable for insurance pricing and urban planning, while future work should address end-to-end fireproof classification, enhanced data augmentation, and incorporating street-level data to further reduce error propagation.

Abstract

Structural fireproof classification is vital for disaster risk assessment and insurance pricing in Japan. However, key building metadata such as construction year and structure type are often missing or outdated, particularly in the second-hand housing market. This study proposes a multi-task learning model that predicts these attributes from facade images. The model jointly estimates the construction year, building structure, and property type, from which the structural fireproof class - defined as H (non-fireproof), T (semi-fireproof), or M (fireproof) - is derived via a rule-based mapping based on official insurance criteria. We trained and evaluated the model using a large-scale dataset of Japanese residential images, applying rigorous filtering and deduplication. The model achieved high accuracy in construction-year regression and robust classification across imbalanced categories. Qualitative analyses show that it captures visual cues related to building age and materials. Our approach demonstrates the feasibility of scalable, interpretable, image-based risk-profiling systems, offering potential applications in insurance, urban planning, and disaster preparedness.

Figures (5)

• Figure 1: Representative facade samples from five combinations of property type and building structure. No sample was available for the combination of No-communal and Concrete-like.
• Figure 2: Distribution of construction years in the filtered dataset. The data is skewed toward more recent buildings.
• Figure 3: Intermediate attributes prediction results
• Figure 4: Examples of facade images where the construction year was predicted within $\pm 3$ years. Each sample corresponds to a different fireproof structural class.
• Figure 5: Examples of misclassified facade images. Each caption shows the predicted and true intermediate attributes (building structure, property type), as well as the resulting structural fireproof classification. Misclassifications often arise due to visually ambiguous features between steel and wooden structures, or between communal and non-communal settings.