insights The Algorithm
We use a Context-Aware Weighted KNN model ($S$) to rank comparable properties. The similarity score (0.0 - 1.0) is calculated as:
$$ S = 0.4(Loc) + 0.3(Area) + 0.2(Beds) + 0.1(Recency) $$
- Location (0.4): 1.0 if same building, 0.6 if same district, 0.2 otherwise.
- Area (0.3): Penalized linearly by percentage difference.
- Bedrooms (0.2): Penalized by 0.5 for each room difference.
Power Law Scaling:
Rent is not linear. A 100m² apartment is not double the price of a 50m² one. We adjust comparable prices ($P_{adj}$) using a diminishing returns formula ($Area^{0.8}$):
$$ P_{adj} = P_{match} \times (\frac{Area_{target}}{Area_{match}})^{0.8} $$
grade Confidence Scores
Confidence is determined by the score of the single best comparable property found.
A
> 85% Match: Identical unit type in same building found.
B
> 75% Match: Different size in same building, or identical in same district.
C
> 65% Match: Similar features in same district.
D
Fallback: No direct matches. Uses global median price/m² for bedroom count.
calculate ROI Formula
(Predicted Rent × 11) ÷ Sale Price
*We multiply by 11 (not 12) to conservatively account for vacancies, management fees, and maintenance costs.