Trained on 480,000+ verified multifamily transactions across the U.S., continuously retrained as new sales close
Machine Learning & AI
Our ensemble architecture combines Random Forest and Gradient Boosting models (50/50 split) with a sophisticated two-model confidence system achieving 0.75 AUC. Random Forest builds hundreds of independent decision trees, each trained on different subsets of data, then averages their predictions to reduce overfitting and improve stability—particularly valuable for handling the diverse property types and market conditions in NYC multifamily. Gradient Boosting takes a sequential approach, where each new tree corrects the errors of previous trees, creating a powerful model that excels at capturing complex, non-linear relationships between property features and sale prices. We employ 95/5 stratified split validation by borough and year, analyzing 40 engineered features with interaction terms to capture complex market dynamics.
Valuation Methodology
Using hedonic pricing principles, we analyze each property characteristic individually to determine its marginal contribution to value. Our system incorporates 200+ neighborhood-specific cap rates derived from actual transaction data, performs implied value analysis across normal, distressed, and development scenarios, and applies market timing adjustments for pre-COVID, COVID, and recovery periods.
Intelligent Detection Systems
Advanced algorithms identify distressed properties using violation data and market signals, recognize development sites by distinguishing land plays from income-generating properties, filter non-market transactions such as family transfers and bulk sales, and segment properties into high, medium, and low quality tiers for more accurate comparisons.
Data Infrastructure
Our training dataset spans 88,639 transactions across multiple market cycles, integrating data from PLUTO, HUD, DOF, and violation databases. Geospatial analysis identifies luxury zones and institutional corridors, while real-time processing delivers instant valuations. The system continuously learns from new transactions to maintain accuracy across changing market conditions.
