AKI Predict
DEU HOSPITAL · RESEARCH
DEU Hospital · ICU Cohort · Research Tool

AKI Mortality
Risk Predictor

A clinical research tool combining four machine learning models to estimate in-hospital mortality risk for Acute Kidney Injury patients in intensive care units.

Research purposes only · Not for clinical decision-making

Scroll
0+
ICU Patients
Retrospective cohort
0
ML Models
LR · RF · GB · ANN
0%
Best AUC
Gradient Boosting
0
Clinical Features
Lab + demographics
Model Performance

Four models. All above 86% AUC.

Each model was independently evaluated on a held-out 20% test set. The best model is selected automatically for predictions.

GB
Gradient Boosting
Area Under ROC Curve
0.0%
RF
Random Forest
Area Under ROC Curve
0.0%
ANN
Neural Network
Area Under ROC Curve
0.0%
LR
Logistic Regression
Area Under ROC Curve
0.0%
Capabilities

Built for clinical research

Every aspect of the system is designed with clinical accuracy, transparency, and reproducibility in mind.

Multi-Model Ensemble

Four independent ML algorithms — Logistic Regression, Random Forest, Gradient Boosting, and ANN — evaluated together for robust predictions.

Clinical Laboratory Values

Analyzes key biomarkers: Creatinine, WBC, Platelet count, Total Bilirubin, AST, INR, eGFR, APTT, BUN, and more.

Instant Risk Classification

Mortality probability and risk class (Low / Moderate / High) calculated in milliseconds from your patient values.

MICE Imputation

Missing clinical values handled automatically via MICEFOREST — 5-iteration multiple imputation trained exclusively on the training set.

Explainable AI

Feature importance via permutation analysis and Gini scores. Know which biomarker drives each prediction.

DEU Hospital Dataset

Trained on a real-world retrospective ICU cohort from Dokuz Eylül University Hospital — not synthetic data.

Methodology

Rigorous end-to-end ML pipeline

From raw PostgreSQL data to production predictions — every step is reproducible, leakage-free, and validated on a stratified hold-out test set.

Read full methodology
Data Loading
PostgreSQL · deu_retro_clean
1
Preprocessing
Encoding · ID removal · EDA
2
Stratified Split
80% train · 20% test · seed=42
3
MICE Imputation
5 iterations · fit on train
4
Model Training
LR · RF · GB · ANN(MLP)
5
Evaluation
AUC · F1 · MCC · Precision
6

Try the risk calculator

Enter five clinical laboratory values and receive an AI-generated mortality probability with risk classification in seconds.

Open Calculator

Research use only · Not for clinical decision-making

AKI Mortality Predictor · DEU Hospital Research · 2024
Research purposes only — not for clinical use