VIEWAPP – a new level: Neural analysis of the quality and class of repairs in apartments
Assessing the condition of an apartment is a key element in calculating the value of collateral. The class of repairs affects the estimated value of the property, which banks rely on when making mortgage decisions, and insurance companies when calculating rates and settlements. Until now, such assessments were carried out manually and depended heavily on the expertise of the specialist. VIEWAPP offers a new approach — automatic classification of repair quality using a neural network trained on real images.
How the neural network works
The new model determines the level of finish in eight classes, from ‘unfinished’ to ‘designer renovation.’ In addition to the main label, it returns a probability distribution across all classes, allowing for flexible analysis of borderline cases. For example, if there is a high probability of two categories — C1 (standard+) and D0 (European-style renovation) — the expert can make a decision based on the context and documentation.
Renovation classes:
- A0 — no finishing
- A1 — white box
- B0 — economy
- B1 — economy+
- C0 — standard
- C1 — standard+
- D0 — European-style renovation
- D1 — designer
The class names and descriptions were refined and approved in collaboration with market representatives.
Automated analysis — a step towards accuracy
The implementation of this neural network opens up the possibility of automating one of the most labour-intensive stages of property valuation — the analysis of property photographs.
This is particularly relevant for mass checks in mortgage, insurance and resale transactions.
System:
- Eliminates subjectivity in assessing the level of finishing;
- Allows banks and insurance companies to standardise their approach to repair classification;
- Simplifies mass portfolio analytics and speeds up decision-making processes.
Impact on collateral value calculation
The repair class is one of the factors affecting the final valuation of a property. For example:
- An apartment with a C1 (standard+) class is valued higher than a similar apartment with a B1 (economy+) class, all other things being equal;
- For new buildings without finishing, the difference between A1 (white box) and B0 (economy) is also critical, especially when banks conduct internal collateral assessments.
Automatic classification allows you to set formal calculation rules and embed them in digital processes, reducing the risk of errors and improving the manageability of the collateral base.
Current status
The neural network is deployed on the server and accessible via API and a special bot for testing. The model has already been retrained and technical metrics have been verified. In the near future, it will be launched on real cases — first in a test set format with manual verification of results. After that, we plan to implement a feedback form and accumulate statistics to further improve accuracy.
The model works quickly (1–2 seconds per photo) and can be easily adapted to the requirements of a specific company, including different approaches to classification or linking classes to price ranges.