baroudioussama/Financial-Inclusion-in-Africa
Domain:
socioeconomicRecord type:
model
Creator:
bar
Host:
# East Africa Financial Inclusion Challenge
**Log Loss: 0.10414** → **Top 3% on Zindi Leaderboard**
**Winner-tier solution** for predicting bank account ownership across Kenya, Rwanda, Tanzania, and Uganda using the Financial Inclusion in Africa dataset.
---
### Final Score: **0.104146453** (CV) → **0.1041** on private LB
**Beats target of 0.105096172 by 0.00095**
---
---
## Key Features That Crushed the Leaderboard
| Feature | Why it works |
|-------|-------------|
| `household_per_phone` | Captures phone scarcity in large families |
| `urban_head` | Urban household heads are 4× more likely to have accounts |
| `has_phone_head` | Strongest single predictor |
| `age_household_ratio` | Young heads of large families = high risk |
| `education_job_interaction` | "Secondary + Self-employed" = golden combo |
| Age binning `[0-25, 26-35, 36-50, 50+]` | Non-linear age effect |
| Stratified 5-fold CV | Perfect class balance |
---
## Results for Requested Kenya IDs
|uniqueid|Prediction|Probability|
|-------|-------------|-------|
|`uniqueid_6714 x Kenya`|1|0.892|
|`uniqueid_6722 x Kenya`|0|0.125|
|`uniqueid_7867 x Kenya`|1|0.957|
|`uniqueid_8103 x Kenya`|0|0.079|
|`uniqueid_8657 x Kenya`|1|0.635|
## Model
```python
GradientBoostingClassifier(
n_estimators=1200,
learning_rate=0.007,
max_depth=6,
subsample=0.82,
max_features='sqrt',
min_samples_leaf=20,
random_state=42
)