hyvioq-v1.0.0 is an XGBoost regression model that estimates flood probability from 20 environmental, infrastructure, climate, population, and disaster-preparedness factors. It is the first released model checkpoint from hyvioq, an ongoing flood-prediction project that also includes a companion mobile application.
The model outputs a continuous FloodProbability value in the approximate range 0–1, interpretable as an estimated probability of flooding under the given input conditions.
This model was independently developed and trained by Bindupautra Jyotibrat.
Model Details
Model Description
hyvioq-v1.0.0 is trained on a large tabular flood-risk dataset and learns the relationship between 20 numerical risk factors and the target variable FloodProbability.
It is intended as a research and demonstration model for flood-risk estimation, not as a substitute for official flood forecasting systems, hydrological models, emergency-management systems, or professional risk assessments.
Support research and educational work involving tabular flood prediction.
Input Features
The model accepts 20 numerical features, each on a scale of approximately 1–10:
#
Feature
#
Feature
1
MonsoonIntensity
11
IneffectiveDisasterPreparedness
2
TopographyDrainage
12
DrainageSystems
3
RiverManagement
13
CoastalVulnerability
4
Deforestation
14
Landslides
5
Urbanization
15
Watersheds
6
ClimateChange
16
DeterioratingInfrastructure
7
DamsQuality
17
PopulationScore
8
Siltation
18
WetlandLoss
9
AgriculturalPractices
19
InadequatePlanning
10
Encroachments
20
PoliticalFactors
Target:FloodProbability
Downstream Use
The model may be integrated into research prototypes, educational applications, data-analysis pipelines, environmental-risk dashboards, or experimental decision-support systems — including as one component of the broader hyvioq mobile app. Any downstream application should clearly communicate that predictions are model estimates, not authoritative flood warnings.
Out-of-Scope Use
This model should not be used as the sole basis for:
Emergency evacuation or public safety decisions
Disaster-response or infrastructure-safety decisions
Insurance, financial, or government policy decisions
Real-time flood forecasting
The model does not ingest real-time rainfall, river-level, weather-radar, satellite, hydrological, or sensor data, so its predictions should not be treated as real-time forecasts.
Bias, Risks, and Limitations
The model is trained on one tabular dataset and may not generalize across all geographic regions or environmental conditions.
Input features are abstract numerical scores rather than direct physical measurements.
The model does not capture temporal dynamics in weather, rainfall, river discharge, or soil moisture.
Predictions may be unreliable for inputs outside the training distribution.
A high (or low) predicted probability does not guarantee that flooding will (or will not) occur.
The model offers no causal explanation for its predictions.
Reported metrics reflect performance on the held-out test split of the training dataset, not real-world deployment.
Recommendations
Use the same feature definitions and numerical ranges as training.
Validate inputs before inference and avoid extrapolating far outside the training distribution.
Treat outputs as estimates, not definitive warnings.
Validate against geographically and temporally independent data before any real-world use.
Combine with real-time meteorological, hydrological, and sensor data for operational systems.
Defer to official governmental flood-warning sources for real-world safety decisions.
How to Get Started with the Model
pip install -r requirements.txt
python
1import xgboost as xgb
23model = xgb.XGBRegressor()4model.load_model("flood_model.json")
Alternatively, run the provided app.py for a local inference interface.
Example Run
Below is an example console session showing values entered for each feature and the resulting prediction and risk level:
Example hyvioq-v1.0.0 inference run
(Replace path/to/inference-example.png with the image path once it's added to the repository, e.g. assets/inference-example.png.)
Training Details
Training Data
The model was trained on the Flood Prediction Dataset, hosted on Kaggle under the CC0 1.0 Universal (Public Domain) license, downloaded via the Kaggle API. train.csv (1,117,957 samples) was used for training and flood.csv for evaluation.
Note: The Kaggle listing notes the dataset has no verifiable external source and is re-uploaded for educational and research purposes only; provenance, collection methods, and exact geographic coverage are unknown.
Training Procedure
Load the dataset with pandas; drop the id column when present.
Separate FloodProbability as the regression target.
Use the remaining 20 columns as model features directly (no neural-network feature extraction or transfer learning).
The notebook also contains an earlier 500-estimator configuration evaluated on flood.csv.
Speeds, Sizes, Times
Training ran on a Google Colab NVIDIA T4 GPU in under one hour.
Training regime: FP32 / standard XGBoost numerical computation
Evaluation
Testing Data
Evaluation used flood.csv from the same dataset, with the same 20 features and FloodProbability target.
Metrics
RMSE — average magnitude of prediction error (lower is better)
R² — proportion of variance explained (closer to 1 is better)
Results
Metric
Score
RMSE
0.01184
R²
0.944
Full evaluation logs and results are available at Results/XGBoost. These results are from the held-out evaluation split and should not be interpreted as guaranteed real-world performance.
Model Examination
The model is a gradient-boosted decision-tree ensemble (XGBoost). It can be examined further with feature importance, SHAP values, or partial-dependence analysis; no formal interpretability study is included in this release.
