Predicting Residential Rents in Dakar Using Machine Learning
cs.AI
Quick verdict
Claims a first interpretable ML rent model for Dakar (XGB R²=0.847); candid on CV drop and 64.2% coverage, but headline is single split, data on request.
Summary
The paper builds what the author presents as the first interpretable machine-learning pipeline for predicting long-term residential rents in Dakar, Senegal. From 1,654 web-scraped listings (January 2024–June 2025), a documented seven-step cleaning pipeline yields 1,507 listings across 24 localities with 33 variables, enriched with engineered features (luxury_score, kw_score). After a log(1+price) transform and leakage-free KFold target encoding of location, five models are compared. Optimized XGBoost (Optuna, 100 trials) performs best on a single 80/20 split (R²=0.847, MAE 210,902 FCFA, RMSE 324,195 FCFA), slightly ahead of tuned LightGBM (R²=0.842). A gain-versus-SHAP divergence is highlighted: location ranks 6th by gain (3.4%) but 2nd by mean |SHAP| (0.155), which the author argues matters for hedonic studies using target-encoded categoricals. Quantile regression (q10–q90) achieves only 64.2% empirical coverage against an 80% nominal target, and 5-fold CV on the original FCFA scale yields a lower mean R² of 0.782 — gaps the author reports candidly. Main limitations: asking rather than transacted prices, selection bias toward online upscale listings, geographic imbalance, no geospatial coordinates, small test set (302 listings), and code/data available only upon request.
Review complete
Assessment at a glance
5
risk flags identified
● high 0
● medium 3
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- An Optuna-optimized XGBoost achieves R²=0.847, MAE 210,902 FCFA and RMSE 324,195 FCFA on a held-out 20% split of 1,507 Dakar rental listings.
- Location's importance flips from 6th (3.4% of gain) to 2nd (mean |SHAP| 0.155) depending on the importance measure, a methodological caution for hedonic studies using target-encoded categoricals.
- Leakage-free KFold target encoding of 24 localities allows fair model comparison without optimistic bias.
- Quantile-regression prediction intervals cover only 64.2% of the test set against an 80% nominal target and require conformal recalibration before operational use.
- The 1,507-listing corpus is an original, documented resource for analyzing Dakar's rental market.
Strengths
Original regional dataset with documented construction
1,654 scraped listings (Jan 2024–Jun 2025) reduced via a seven-step cleaning pipeline (§3.2) to 1,507 listings, 24 localities, 130+ raw locality labels normalized; the first documented long-term rental corpus for Dakar.
Candid negative reporting
The author himself contrasts the single-split headline (R²=0.847) with CV on the FCFA scale (mean R²=0.782±0.034, RMSE 369,744; §4.6) and reports quantile interval undercoverage of 64.2% vs 80% nominal (§4.9, §5.5) instead of hiding them.
Leakage-aware preprocessing
Transferable interpretability finding
Fully specified experimental setup
Weaknesses
Headline rests on the favorable single split
No uncertainty on model comparisons
Correlated features confound the interpretability claims
Not independently reproducible
Internal inconsistencies
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Preliminary Review: Predicting Residential Rents in Dakar Using Machine Learning
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 5 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
How accurate is the rent prediction model for Dakar?
On a held-out 20% test split, the tuned XGBoost model reaches R²=0.847 with a mean absolute error of about 210,900 FCFA per month. However, 5-fold cross-validation on the original FCFA scale gives a lower mean R² of 0.782 and RMSE of about 369,700 FCFA, so roughly 0.78 is the more robust accuracy estimate.
Is the Dakar rental dataset public?
No. The paper states the 1,507-listing dataset (labeled CC BY 4.0) and the notebook are available only upon request from the author, not in a public repository, so third parties cannot currently verify the results.
Why do gain importance and SHAP disagree about location?
The target-encoded location variable is continuous and appears in many splits that each yield small loss reductions, so its average gain per split looks small (3.4%, rank 6). SHAP instead accumulates each observation's total contribution, ranking it 2nd (mean absolute value 0.155). The paper argues SHAP better reflects actual influence, though correlated duplicate features complicate the comparison.
What are typical rents in Dakar according to this study?
In the scraped corpus (mid-2024 to mid-2025 listings), the median asking rent is 700,000 FCFA per month (mean 950,411, standard deviation 796,534), and the model's base predicted level is about 686,870 FCFA. These are asking prices, not negotiated transaction rents.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 2–9 appear only as embedded images/captions in the extraction; visual claims (e.g., monotone SHAP color patterns, residual dispersion) rest on the author's textual descriptions.
- The SHAP waterfall example (§4.8) is described but its numerical decomposition exists only in the accompanying notebook, not in the text.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.