What you are looking at
What
The engineered feature set used to train all 5 ML models —
one row per property (1,011 total). Features include geographic flags (coastal, urban, game reserve, wine region),
a data-quality score, the SA season at scrape date, and the K-Means cluster assignment.
Why
Raw price and review counts alone are weak predictors. Adding location context
(coastal = premium uplift), property type signals (game lodge = luxury segment),
and seasonal timing significantly improves model accuracy.
Feature engineering is where most ML value is created.
So What
The model cards at the top show how well each model performs.
R² = 0.82 on demand prediction means the model explains 82% of demand variance
across properties. The feature table shows which cluster each property belongs to —
use cluster labels to quickly segment the market.
How to read the model cards: R² (R-squared) = % of variance the model explains — 0.82 means 82% of demand variation is explained by the features. F1 = balance of precision and recall for classification — 1.0 = perfect. AUC = ability to rank positive vs negative outcomes — 0.5 = random chance, 1.0 = perfect. All models were trained on synthetic data — metrics are illustrative.
Demand Predictor
Random Forest
R² = 0.82
Price Classifier
Gradient Boosting
F1 = 0.79
Conversion Predictor
Logistic Regression
AUC = 0.74
Anomaly Detector
Isolation Forest
Precision = 0.88
Regional Clustering
K-Means k=4
Silhouette = 0.61
Feature Summary
179
Coastal Properties
94
Game Lodge / Safari
79
Wine Region
69
High-Demand Cluster
797
Emerging Gems
1011
Total Properties
Feature Table (sample 50, sorted by demand)
How to read the K-Means Cluster labels:
Clusters were found by grouping all 114 regions using K-Means (k=4) on four features: average nightly price, average demand score, average review count, and listing count. The algorithm finds natural groupings — these labels are descriptive names assigned after inspecting the centroid values of each group.
High-Demand Hotspot — 7 regions. Cheapest avg price (R1,297) but highest demand (24.04). These are volume-driven, budget-accessible destinations where supply cannot keep up with demand. Cape Town suburbs, popular Garden Route towns.
Established Premium — 19 regions. Mid-price (R1,651), strong reviews, steady demand. These markets are healthy and consistent year-round. Safe investment destinations.
Value Volume Leader — 83 regions (the majority). Higher avg price (R1,851) but mid demand. Wide variety; many mid-market properties. Price sensitivity is real in this cluster.
Emerging Gem — 5 ultra-premium regions (avg R9,754). Very low listing count, very low demand. These are niche luxury destinations. High barrier to entry but low competition.
Clusters were found by grouping all 114 regions using K-Means (k=4) on four features: average nightly price, average demand score, average review count, and listing count. The algorithm finds natural groupings — these labels are descriptive names assigned after inspecting the centroid values of each group.
High-Demand Hotspot — 7 regions. Cheapest avg price (R1,297) but highest demand (24.04). These are volume-driven, budget-accessible destinations where supply cannot keep up with demand. Cape Town suburbs, popular Garden Route towns.
Established Premium — 19 regions. Mid-price (R1,651), strong reviews, steady demand. These markets are healthy and consistent year-round. Safe investment destinations.
Value Volume Leader — 83 regions (the majority). Higher avg price (R1,851) but mid demand. Wide variety; many mid-market properties. Price sensitivity is real in this cluster.
Emerging Gem — 5 ultra-premium regions (avg R9,754). Very low listing count, very low demand. These are niche luxury destinations. High barrier to entry but low competition.
| Property | Region | TypeAccommodation category from LekkeSlaap listing | Price (R)Nightly rate in South African Rand (ZAR) at scrape date | ReviewsTotal cumulative guest reviews | DemandComposite 0-100. Higher = more in-demand than average | CoastalWithin 50km of coastline | GameGame reserve / safari listing or region | WineCape Winelands region | QualityData completeness 0-1. 1.0 = all key fields present | ClusterK-Means market segment. See explainer above. |
|---|