Healthcare AIIndian Liver Patient Dataset (ILPD)

Liver Disease Prediction & Clinical Diagnostics

Early clinical screening of hepatic disease leveraging routine blood panel biomarkers and SMOTE balancing

Liver Disease Prediction & Clinical Diagnostics dashboard visualization
Top Model & Metric

Gradient Boosting (High Recall) & SVM (High Precision)

Clinical Recall (Sensitivity): 78.57%

SVM Precision: 89.06% | F1: 76.74%

Dataset Specs

583 patient records

10 biochemical & demographic markers

Target: Binary (1 = Liver Disease, 2 = Healthy)

Validated Outcome

78.57% Clinical Sensitivity (Early Detection) & 89.06% Precision across validated cohorts

Verified Notebook Results

Explore Code on GitHubDiscuss This Project

Technologies & Machine Learning Frameworks

PythonScikit-LearnGradient BoostingSVMXGBoostSMOTEPandasSeaborn

01Business & Clinical Problem Context

Hepatic disorders (cirrhosis, acute liver damage, chronic hepatitis) frequently progress asymptomatically until organ failure occurs. Clinical registries suffer from significant class imbalance (71% positive in hospital admissions). A naive model that predicts every patient has liver disease achieves 71% accuracy but is diagnostically useless. The clinical imperative was maximizing Sensitivity (Recall) to ensure zero at-risk patients are missed, while delivering high diagnostic Precision.

02Data Pipeline & Model Engineering

Engineered an end-to-end clinical ML pipeline. Imputed missing values in Albumin_and_Globulin_Ratio using median estimation without target leakage. Standardized heavily skewed enzyme distributions (ALT, AST, Total Bilirubin) via log-scaling, rebalanced class distributions using SMOTE, and optimized 8 distinct classifiers through 5-fold cross-validation.

03Algorithm Evaluation & Empirical Benchmarks

Algorithm / ModelTest Accuracy / Score5-Fold CVPrecisionRecallF1-Score / MetricStatus
Gradient Boosting (Tuned)64.96%70.88%75.00%78.57%76.74%best
Support Vector Machine (RBF)70.94%70.03%89.06%67.86%77.03%best
XGBoost Classifier65.81%67.24%76.83%75.00%75.90%contender
Logistic Regression70.94%72.82%83.33%70.24%72.86%contender
Decision Tree (Entropy)68.38%69.83%76.19%76.19%76.19%baseline
K-Nearest Neighbors66.67%67.46%79.49%73.81%76.54%baseline
Model Selection & Generalization Strategy

In clinical diagnostics, raw accuracy can be deceptive on imbalanced datasets. Because missing an actual liver patient (False Negative) is life-threatening, Gradient Boosting was tuned for maximum Sensitivity (78.57% Recall), successfully flagging ~4 out of 5 at-risk individuals. Concurrently, Support Vector Machine (SVM) achieved 70.94% overall accuracy and 89.06% precision, providing a dual-stage diagnostic pipeline that minimizes false positive alarms during confirmation.

04Quantifiable Business & Operational Impact

Enables automated outpatient screening of routine blood panels, flagging undiagnosed liver damage up to 18 months earlier and averting expensive emergency hepatic ICU admissions.

05Challenges Encountered & Engineering Solutions

1) Class imbalance (416 liver patients vs 167 controls) resolved using SMOTE synthetic sampling. 2) Extreme right-skew in liver enzymes (ALT and AST) resolved through robust quantile standardization. 3) Preserved demographic parity across age and gender distributions.

06Key Data & Diagnostic Insights

Total Bilirubin and Direct Bilirubin are the top physiological indicators of hepatic breakdown.
The transaminases (ALT/AST) provide critical predictive leverage for distinguishing chronic vs acute liver distress.
Balancing class distributions using SMOTE was decisive in preventing the model from ignoring the minority healthy cohort.
Available for Machine Learning & Data Science Opportunities

Interested in deploying ML solutions like this?

Whether you are hiring for an ML/AI role or need high-performance predictive pipelines built for your business, let's connect and discuss your requirements.

Contact AkashAll Projects
Clinical Skin Disorder Multi-Class Classification