A data mining Project that applies classification and clustering techniques to predict liver disease using patient medical records — helping identify risk factors and support early diagnosis.
Liver disease is a growing global health concern, affecting people of all ages and nationalities and causing millions of deaths annually. This project was initiated to understand the causes, analyze contributing factors, explore prevention methods, and extract meaningful insights from real patient data.
| Name | ID |
|---|---|
| Raghad Fares | 443200793 |
| Sereen Al-hmoud | 443200463 |
| Aeshah Almakhlifi | 443200713 |
| Luluh Al-yahya | 443200609 |
Course: IT 326 – Data Mining · Lab: Wednesday · Section: 74557 Institution: King Saud University – College of Computer and Information Sciences
| Property | Details |
|---|---|
| Source | Kaggle – Liver Disorders |
| Records | 583 patients |
| Attributes | 11 (Age, Gender, TB, DB, Alkphos, Sgpt, Sgot, TP, ALB, A/G Ratio, Selector) |
| Target | Selector — 1 = Liver Disease · 2 = No Liver Disease |
| Missing Values | 4 missing values in A/G Ratio column |
Data Understanding → Preprocessing → Data Mining → Evaluation → Findings
Key statistical observations:
- Age ranges 4–90 years — liver disease spans all age groups
- TB, DB, Alkphos, Sgpt, Sgot — very high variance, extreme values present
- TP, ALB, A/G Ratio — low to moderate variance, more stable indicators
Visualizations used: Pie Charts, Scatter Plots, Box Plots, Line Graphs
| Step | Method |
|---|---|
| Missing Values | Replaced with column mean (A/G Ratio) |
| Outlier Handling | IQR capping — 388 rows affected |
| Encoding | Gender: Male = 1, Female = 0 |
| Normalization | Min-Max scaling to unify attribute ranges |
| Discretization | Age → Children (0–17), Adults (18–64), Seniors (65+) |
| Aggregation | Grouped by Gender + Selector to analyze mean differences |
| Class Balancing | Downsampled majority class to achieve 40–60% balance |
Predicts whether a patient has liver disease based on 10 medical attributes.
| Split | Criterion | Accuracy | Precision | Sensitivity | Specificity | Error Rate |
|---|---|---|---|---|---|---|
| 70/30 | Info Gain | 65.0% | 63% | 48% | 78% | 34% |
| 60/40 | Info Gain | 62.5% | 56% | 47% | 73% | 37.4% |
| 80/20 | Info Gain | 62.1% | 50% | 38% | 76% | 37.8% |
| 70/30 | Gini Index | 64.0% | 61% | 48% | 76% | 35% |
| 60/40 | Gini Index | 63.0% | 57.3% | 51% | 72% | 36% |
| 80/20 | Gini Index ✅ | 67.0% | 57.6% | 48% | 78% | 32% |
Best Model: 80/20 split with Gini Index — highest accuracy (67%) and lowest error rate (32%)
The decision tree relies primarily on: Total Bilirubin (TB), followed by Sgot, Alkphos, ALB, TP, DB, and Sgpt.
Groups patients by similarity without using the class label. Tested K = 2, 3, 6.
| K | Avg. Silhouette Score | WSS |
|---|---|---|
| K=2 ✅ | 0.329 | 2537.0 |
| K=3 | 0.232 | 2125.6 |
| K=6 | 0.243 | 1526.8 |
Best Clustering: K=2 — highest silhouette score, most distinct and cohesive clusters
Validation: Silhouette Method + Elbow (WSS) Method
- Men are significantly more susceptible to liver disease than women
- No strong age correlation — disease can affect individuals of all ages
- Elevated TB, DB, Alkphos, Sgpt, Sgot strongly associated with liver disease
- Higher ALB (Albumin) indicates liver health — lower levels suggest disease
- A decreasing A/G Ratio may signal declining liver function
- Classification outperforms clustering here since the dataset includes known labels
- Fatemeh Mehrparvar, Liver Disorders Dataset, Kaggle
- Liver disease accounts for two million annual deaths globally, PubMed
- How Many People Have Liver Disease?, American Liver Foundation
- Labs and Lecture Slides, IT Department, King Saud University