⚡ Advanced ML & Geospatial Intelligence

Road Accident Severity Predictor

Real-time machine learning inference and geospatial intelligence for traffic incident injury prediction.

Balanced Accuracy

82.4% +14.2%

Macro F1 Score

0.801 SMOTE

Inference Latency

< 8ms Real-Time

Spatial Engine

DBSCAN Haversine

Roadway & Speed Dynamics

Kinetic Multiplier

Posted Speed Limit 45 mph

Trafficway Separation

Intersection Zone

Visibility & Friction

Weather Condition

Lighting Environment

Incident Time of Day 19:00

Surface Traction

Vehicle & Damage

First Collision Type

Estimated Damage Value

Primary Contributing Cause

Total Vehicles Involved 2 Units

High Severity Incident

84%

Injury Prob.

High risk of passenger injury or disabling vehicle structural damage requiring immediate emergency towing.

Injury & Tow Probability 84.2%

No Injury / Drive Away 15.8%

Calculated via game-theoretic Shapley Additive Explanations across test validation splits.

TreeExplainer

DBSCAN density clustering with Haversine distance and severity-weighted risk indices.

Postcode Corridors

Cluster ID	Postal Code / Corridor	Reported Volume	Severity Index	Priority Tier
#C-01	60607 (Loop / I-90 Junction)	1,420 crashes	8.92 / 10.0	Critical Priority
#C-02	60616 (Near South Side / Cermak)	1,150 crashes	8.45 / 10.0	Critical Priority
#C-03	60618 (Avondale / Elston Corridor)	820 crashes	6.30 / 10.0	Elevated
#C-04	60647 (Logan Square / Fullerton)	740 crashes	5.85 / 10.0	Elevated
#C-05	60630 (Jefferson Park)	410 crashes	3.20 / 10.0	Standard Baseline

Stratified 5-Fold cross-validation results comparing candidate architectures.

LightGBM v4.7

Algorithm Candidate	Sampling Technique	Balanced Accuracy	Macro F1	ROC-AUC	Status
LightGBM Classifier	SMOTE (1:1 Ratio)	82.4%	0.801	0.875	Production Model
Random Forest (100 Trees)	Random Undersampling	74.1%	0.713	0.790	Evaluated
Logistic Regression	Class Weight Balanced	68.2%	0.650	0.722	Baseline