🧬 Liver Cirrhosis Stage Detection Project

This project focuses on predicting the histologic stage of liver cirrhosis based on various medical and clinical indicators of a patient. Using a dataset from a Mayo Clinic study on primary biliary cirrhosis (PBC) conducted between 1974 and 1984, machine learning models were trained to classify patients into one of three stages of liver damage:

• Stage 1 → Mild
• Stage 2 → Moderate
• Stage 3 → Severe

---

📊 Key Highlights

• Dataset Source:

  • Mayo Clinic study on Primary Biliary Cirrhosis (PBC).
  • Period: 1974–1984.

• Target Variable:

  • Stage of cirrhosis (1, 2, or 3).

• Features Used:
  The dataset includes medical and biochemical attributes such as:

  • N_Days: Duration from registration to death/transplant/study end
  • Status: Patient status – C (Censored), CL (Censored due to liver transplant), D (Death)
  • Drug: Drug administered – D-penicillamine or placebo
  • Age: Age in days
  • Sex: M or F
  • Ascites, Hepatomegaly, Spiders: Presence of medical symptoms (Y/N)
  • Edema: Level of edema severity (N/S/Y)
  • Biochemical values:
    • Bilirubin (mg/dl), Cholesterol (mg/dl), Albumin (gm/dl), Copper (ug/day)
    • Alk_Phos (U/l), SGOT (U/ml), Triglycerides (mg/dl)
    • Platelets (per 1000 ml), Prothrombin time (sec)

---

🛠 Feature Engineering

• Converted categorical features (e.g., Sex, Drug, Edema, etc.) to numerical format.
• Handled missing data using appropriate imputation strategies.
• Normalized numerical features for improved model performance.

---

🤖 Models Evaluated

A range of classification models were evaluated using StackingClassifier with different meta (final) estimators:

• Models Tested:

  • Logistic Regression → 59.64% accuracy
  • SGD Classifier → 58.38% accuracy
  • LDA → 59.40% accuracy
  • Random Forest → 94.76% accuracy
  • XGBoost → 95.54% accuracy
  • SVM → 84.16% accuracy
  • KNN → 88.42% accuracy
  • Gaussian → 50.54% accuracy

• Base Learners:

  • RandomForestClassifier
  • XGBClassifier

• Final Estimators Tested:

  • SVC → 95.90% accuracy
  • LogisticRegression → 95.74% accuracy
  • SGDClassifier → 95.64% accuracy
  • LinearDiscriminantAnalysis → 95.80% accuracy
  • RandomForestClassifier → 95.86% accuracy
  • XGBClassifier → 96.14% accuracy

• Final Estimator:

  • XGBClassifier was chosen as the final meta-model due to its superior performance.

---

🧪 Prediction Results

• Test 1:

  • Input: Synthetic random patient data
  • Output:
    • Predicted Cirrhosis Stage: 1 → Stage 1 (Mild)
  • ✔️ Shows model is capable of detecting low-risk cases.

• Test 2:

  • Input: Real sample from original dataset
  • Output:
    • Predicted Cirrhosis Stage: 3 → Stage 3 (Severe)
  • ✔️ Confirms model can detect advanced liver damage from clinical data.

---

📌 Usage

• Input patient features (age, lab results, symptoms, etc.) in the required format.
• The model will output the predicted stage of cirrhosis with a corresponding severity label.
• Can assist medical professionals in early detection and prioritization of treatment.

Python Version

• Python 3.12.5