🧪 Multiclass Classification of Liver Disease
🩺 Project Overview This repository contains a clinical analytics pipeline built in R to classify patients into one of four diagnostic categories based on biochemical and demographic markers. The project focuses on transforming raw clinical data into a high-precision diagnostic tool by combining traditional statistical modeling with advanced dimensionality reduction.
The final pipeline utilizes Linear Discriminant Analysis (LDA) on PCA-transformed features to achieve high classification accuracy across diverse liver disease stages.
🚀 Key Features & Methodology
-
🧹 Clinical Data Engineering
Preprocessing: Cleaned and encoded mixed-type demographic and biochemical data.
Outlier Management: Implemented 99th percentile capping to handle extreme clinical values without losing data integrity.
Feature Scaling: Applied Z-score normalization to standardize features for distance-based and variance-based algorithms.
-
📉 Feature Selection & Noise Reduction
Multicollinearity Control: Applied Principal Component Analysis (PCA) to handle highly correlated biochemical markers.
Biomarker Identification: Used ANOVA and effect-size analysis to validate high-signal biomarkers, identifying AST, BIL, and GGT as the most significant drivers of patient segmentation.
-
🤖 Multiclass Modeling & Performance
Models Evaluated: Compared Multinomial Logistic Regression (MLR), Linear Discriminant Analysis (LDA), and Quadratic Discriminant Analysis (QDA).
Champion Model: The LDA model (using PCA features) outperformed others with:
✅ Accuracy: 94.91%
🎯 Macro F1-Score: 0.79
🛠️ Technical Stack Language: R
Libraries: tidyverse, caret, MASS (LDA/QDA), FactoMineR (PCA)
Techniques: PCA, ANOVA, Outlier Capping, Z-score Scaling, Multiclass Classification
📂 Repository Structure scripts/: R scripts for preprocessing, PCA, and model training.
analysis/: ANOVA results and biomarker effect-size reports.
results/: Confusion matrices and accuracy benchmarks.
🏁 How to Use 📥 Load your liver disease dataset into the R environment.
⚙️ Run the preprocessing script to handle outliers and scaling.
📉 Execute the PCA script to generate principal components.
🏆 Run the modeling script to train the LDA classifier and view performance metrics.