Dataset check

Target "flood" ยท 3,908 records

Verdict

Ready to train
  • No blocking problem found. Read the advice below before training.
Advice
  • Strong class imbalance (1:4.8). Use class weights ("balanced"), judge models by PR-AUC / TSS rather than accuracy, and remember probabilities reflect this sampling ratio.
  • The flood label is spatially clustered (Moran I = 0.69, p = 0.005). Random cross-validation would be over-optimistic: use spatial block CV.
  • Suggested spatial block size: about 1,073 m (median autocorrelation range of 3 variable(s); capped at 1,073 to leave enough blocks for 5 folds).
Records
3908
Flood (1)
3237
Non-flood (0)
671

Figures

Class balance and sample locations.
Class balance and sample locations.
Predictor distributions by class.
Predictor distributions by class.
Spearman correlation between predictors.
Spearman correlation between predictors.
Single-predictor separability.
Single-predictor separability.
Missing values.
Missing values.

Predictors

featuretypemissing_pctn_uniqueminmedianmaxsd
distance_to_river1numeric0.17937740.0000.0100.0520.013
distance_to_road1numeric0.17938520.0000.0040.0260.005
elevation_srtm_30m_clippednumeric0.07718041.7825.83313.3211.492
ndvi_s2_10mnumeric0.0003908-0.3790.4820.7390.185
slope_degree_30m_clippednumeric0.07738200.0001.5267.3290.782

Variance-inflation factor

featurevif
distance_to_river11.225
distance_to_road11.221
elevation_srtm_30m_clipped1.162
ndvi_s2_10m1.114
slope_degree_30m_clipped1.050