Add chapter-05-data-mining/README.md
Browse files- chapter-05-data-mining/README.md +391 -0
chapter-05-data-mining/README.md
ADDED
|
@@ -0,0 +1,391 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Chương 5: Thuật toán Khai thác Dữ liệu Cốt lõi
|
| 2 |
+
## Core Data Mining Algorithms
|
| 3 |
+
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
## 📚 Mục tiêu học tập (CLO3, CLO4, CLO5, CLO7, CLO8)
|
| 7 |
+
|
| 8 |
+
Sau khi hoàn thành chương này, sinh viên có thể:
|
| 9 |
+
- Implement thuật toán Apriori và FP-Growth cho Association Rules
|
| 10 |
+
- Xây dựng Decision Tree (ID3, C4.5) cho bài toán Classification
|
| 11 |
+
- Áp dụng Naive Bayes cho phân loại văn bản/dữ liệu
|
| 12 |
+
- Thực hiện K-Means và K-Medoids cho Clustering
|
| 13 |
+
- Sử dụng DBSCAN cho density-based clustering
|
| 14 |
+
- Phân tích outlier bằng phương pháp thống kê và khoảng cách
|
| 15 |
+
|
| 16 |
+
---
|
| 17 |
+
|
| 18 |
+
## 5.1. Khai thác Luật kết hợp (Association Rules)
|
| 19 |
+
|
| 20 |
+
### Các khái niệm cơ bản
|
| 21 |
+
|
| 22 |
+
```
|
| 23 |
+
Transaction Database:
|
| 24 |
+
TID | Items
|
| 25 |
+
─────┼──────────────────────
|
| 26 |
+
T1 | {Bread, Milk, Butter}
|
| 27 |
+
T2 | {Bread, Milk}
|
| 28 |
+
T3 | {Milk, Eggs}
|
| 29 |
+
T4 | {Bread, Milk, Eggs}
|
| 30 |
+
T5 | {Bread, Butter}
|
| 31 |
+
|
| 32 |
+
Itemset: Tập các items (VD: {Bread, Milk})
|
| 33 |
+
|
| 34 |
+
Support(X) = P(X) = count(X) / N
|
| 35 |
+
VD: Support({Bread, Milk}) = 3/5 = 0.6 (60%)
|
| 36 |
+
|
| 37 |
+
Confidence(X → Y) = P(Y|X) = Support(X∪Y) / Support(X)
|
| 38 |
+
VD: Confidence({Bread} → {Milk}) = Support({Bread,Milk}) / Support({Bread})
|
| 39 |
+
= 3/5 / 4/5 = 0.75 (75%)
|
| 40 |
+
|
| 41 |
+
Lift(X → Y) = Confidence(X→Y) / Support(Y)
|
| 42 |
+
VD: Lift({Bread}→{Milk}) = 0.75 / 0.8 = 0.9375
|
| 43 |
+
|
| 44 |
+
Lift > 1: X và Y có quan hệ positive
|
| 45 |
+
Lift = 1: X và Y independent
|
| 46 |
+
Lift < 1: X và Y có quan hệ negative
|
| 47 |
+
```
|
| 48 |
+
|
| 49 |
+
---
|
| 50 |
+
|
| 51 |
+
## 5.2. Thuật toán Apriori
|
| 52 |
+
|
| 53 |
+
### Nguyên lý Apriori
|
| 54 |
+
|
| 55 |
+
> **Nếu một itemset không frequent, thì mọi superset của nó cũng không frequent.**
|
| 56 |
+
>
|
| 57 |
+
> Ngược lại: mọi subset của frequent itemset đều phải frequent.
|
| 58 |
+
|
| 59 |
+
```
|
| 60 |
+
Apriori Algorithm (min_support = 0.4):
|
| 61 |
+
|
| 62 |
+
Bước 1: Frequent 1-itemsets (C1 → L1)
|
| 63 |
+
┌──────────┬──────────┬─────────┐
|
| 64 |
+
│ Itemset │ Support │ Frequent│
|
| 65 |
+
├──────────┼──────────┼─────────┤
|
| 66 |
+
│ {Bread} │ 4/5=0.80 │ ✅ │
|
| 67 |
+
│ {Milk} │ 4/5=0.80 │ ✅ │
|
| 68 |
+
│ {Butter} │ 2/5=0.40 │ ✅ │
|
| 69 |
+
│ {Eggs} │ 2/5=0.40 │ ✅ │
|
| 70 |
+
└──────────┴──────────┴─────────┘
|
| 71 |
+
|
| 72 |
+
Bước 2: Generate candidates C2 từ L1, scan DB
|
| 73 |
+
┌────────────────┬──────────┬─────────┐
|
| 74 |
+
│ Itemset │ Support │ Frequent│
|
| 75 |
+
├────────────────┼──────────┼─────────┤
|
| 76 |
+
│ {Bread, Milk} │ 3/5=0.60 │ ✅ │
|
| 77 |
+
│ {Bread,Butter} │ 2/5=0.40 │ ✅ │
|
| 78 |
+
│ {Bread, Eggs} │ 1/5=0.20 │ ❌ prune│
|
| 79 |
+
│ {Milk, Butter} │ 1/5=0.20 │ ❌ prune│
|
| 80 |
+
│ {Milk, Eggs} │ 2/5=0.40 │ ✅ │
|
| 81 |
+
│ {Butter, Eggs} │ 0/5=0.00 │ ❌ prune│
|
| 82 |
+
└────────────────┴──────────┴─────────┘
|
| 83 |
+
|
| 84 |
+
Bước 3: C3 từ L2 → {Bread, Milk, Butter}?
|
| 85 |
+
Check subsets: {Bread,Butter}✅ {Milk,Butter}❌ → Prune!
|
| 86 |
+
→ L3 = ∅ → STOP
|
| 87 |
+
|
| 88 |
+
Kết quả Frequent Itemsets:
|
| 89 |
+
L1: {Bread}, {Milk}, {Butter}, {Eggs}
|
| 90 |
+
L2: {Bread,Milk}, {Bread,Butter}, {Milk,Eggs}
|
| 91 |
+
```
|
| 92 |
+
|
| 93 |
+
### FP-Growth (cải tiến Apriori)
|
| 94 |
+
|
| 95 |
+
```
|
| 96 |
+
FP-Growth vs Apriori:
|
| 97 |
+
|
| 98 |
+
Apriori: FP-Growth:
|
| 99 |
+
- Scan DB nhiều lần - Scan DB chỉ 2 lần
|
| 100 |
+
- Generate candidates - KHÔNG generate candidates
|
| 101 |
+
- Chậm với large DB - Nhanh hơn nhiều
|
| 102 |
+
- Dễ hiểu - Phức tạp hơn
|
| 103 |
+
|
| 104 |
+
FP-Growth xây dựng FP-Tree (compact representation)
|
| 105 |
+
rồi mine frequent patterns trực tiếp từ tree.
|
| 106 |
+
```
|
| 107 |
+
|
| 108 |
+
---
|
| 109 |
+
|
| 110 |
+
## 5.3. Classification: Decision Trees
|
| 111 |
+
|
| 112 |
+
### Information Gain và Entropy
|
| 113 |
+
|
| 114 |
+
```
|
| 115 |
+
Entropy(S) = -Σ p_i * log2(p_i)
|
| 116 |
+
|
| 117 |
+
VD: Dataset có 9 positive, 5 negative
|
| 118 |
+
Entropy = -9/14 * log2(9/14) - 5/14 * log2(5/14) = 0.940
|
| 119 |
+
|
| 120 |
+
Information Gain(S, A) = Entropy(S) - Σ |Sv|/|S| * Entropy(Sv)
|
| 121 |
+
|
| 122 |
+
Chọn attribute có Information Gain cao nhất để split!
|
| 123 |
+
```
|
| 124 |
+
|
| 125 |
+
### ID3 Algorithm
|
| 126 |
+
|
| 127 |
+
```
|
| 128 |
+
ID3 Decision Tree cho Olist (đơn giản hóa):
|
| 129 |
+
|
| 130 |
+
Target: Is customer satisfied? (review_score >= 4)
|
| 131 |
+
|
| 132 |
+
┌──────────────┐
|
| 133 |
+
│delivery_days │
|
| 134 |
+
│ IG = 0.32 │
|
| 135 |
+
└──────┬───────┘
|
| 136 |
+
│
|
| 137 |
+
┌───────────┼───────────┐
|
| 138 |
+
│ │ │
|
| 139 |
+
≤ 7 days 8-14 days > 14 days
|
| 140 |
+
│ │ │
|
| 141 |
+
┌────▼────┐ ┌────▼────┐ ┌────▼────┐
|
| 142 |
+
│ price │ │Satisfied│ │ NOT │
|
| 143 |
+
│IG=0.15 │ │ (68%) │ │Satisfied│
|
| 144 |
+
└────┬────┘ └─────────┘ │ (82%) │
|
| 145 |
+
│ └─────────┘
|
| 146 |
+
┌────┼────┐
|
| 147 |
+
│ │
|
| 148 |
+
≤ 200 > 200
|
| 149 |
+
│ │
|
| 150 |
+
┌────▼────┐ ┌──▼──────┐
|
| 151 |
+
│Satisfied│ │ Check │
|
| 152 |
+
│ (90%) │ │ freight │
|
| 153 |
+
└─────────┘ └─────────┘
|
| 154 |
+
|
| 155 |
+
ID3: Chỉ dùng categorical (discretize nếu cần)
|
| 156 |
+
C4.5: Hỗ trợ continuous attributes, handling missing values
|
| 157 |
+
```
|
| 158 |
+
|
| 159 |
+
### C4.5 Improvements over ID3
|
| 160 |
+
|
| 161 |
+
| Feature | ID3 | C4.5 |
|
| 162 |
+
|---------|-----|------|
|
| 163 |
+
| Attributes | Categorical only | Categorical + Continuous |
|
| 164 |
+
| Missing values | Not handled | Uses probability weighting |
|
| 165 |
+
| Overfitting | Not addressed | Post-pruning (error-based) |
|
| 166 |
+
| Splitting criterion | Information Gain | **Gain Ratio** = IG / SplitInfo |
|
| 167 |
+
| Multi-way split | Yes | Yes + Binary split for continuous |
|
| 168 |
+
|
| 169 |
+
```
|
| 170 |
+
Gain Ratio = Information_Gain(A) / SplitInformation(A)
|
| 171 |
+
|
| 172 |
+
SplitInfo(A) = -Σ |Sv|/|S| * log2(|Sv|/|S|)
|
| 173 |
+
|
| 174 |
+
→ Penalize attributes with many values (fix IG bias)
|
| 175 |
+
```
|
| 176 |
+
|
| 177 |
+
---
|
| 178 |
+
|
| 179 |
+
## 5.4. Naive Bayes Classification
|
| 180 |
+
|
| 181 |
+
### Bayes' Theorem
|
| 182 |
+
|
| 183 |
+
```
|
| 184 |
+
P(C|X) = P(X|C) * P(C) / P(X)
|
| 185 |
+
|
| 186 |
+
Posterior = Likelihood × Prior / Evidence
|
| 187 |
+
|
| 188 |
+
Naive assumption: features are conditionally independent
|
| 189 |
+
P(X|C) = P(x1|C) × P(x2|C) × ... × P(xn|C)
|
| 190 |
+
|
| 191 |
+
Ví dụ: Predict review_score ≥ 4 (satisfied)
|
| 192 |
+
|
| 193 |
+
Given: delivery_days=5, price=150, state=SP
|
| 194 |
+
|
| 195 |
+
P(satisfied | features) ∝ P(delivery=5|sat) × P(price=150|sat) × P(SP|sat) × P(sat)
|
| 196 |
+
P(unsatisfied | features) ∝ P(delivery=5|unsat) × P(price=150|unsat) × P(SP|unsat) × P(unsat)
|
| 197 |
+
|
| 198 |
+
Chọn class có probability cao hơn.
|
| 199 |
+
```
|
| 200 |
+
|
| 201 |
+
### Types of Naive Bayes
|
| 202 |
+
|
| 203 |
+
| Type | Use case | P(x|C) |
|
| 204 |
+
|------|----------|--------|
|
| 205 |
+
| **Gaussian NB** | Continuous features | Normal distribution |
|
| 206 |
+
| **Multinomial NB** | Text/Count data | Multinomial distribution |
|
| 207 |
+
| **Bernoulli NB** | Binary features | Bernoulli distribution |
|
| 208 |
+
| **Complement NB** | Imbalanced text | Complement of class |
|
| 209 |
+
|
| 210 |
+
---
|
| 211 |
+
|
| 212 |
+
## 5.5. K-Means Clustering
|
| 213 |
+
|
| 214 |
+
### Algorithm
|
| 215 |
+
|
| 216 |
+
```
|
| 217 |
+
K-Means Algorithm:
|
| 218 |
+
|
| 219 |
+
Input: K (số clusters), Data points
|
| 220 |
+
|
| 221 |
+
1. Khởi tạo K centroids (random hoặc K-Means++)
|
| 222 |
+
2. REPEAT:
|
| 223 |
+
a. ASSIGN: Gán mỗi point vào cluster có centroid gần nhất
|
| 224 |
+
b. UPDATE: Tính lại centroid = mean of all points trong cluster
|
| 225 |
+
3. UNTIL centroids không đổi (hoặc max iterations)
|
| 226 |
+
|
| 227 |
+
Iteration 0: Iteration 1: Converged:
|
| 228 |
+
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
|
| 229 |
+
│ * │ │ * . . │ │ * . . │
|
| 230 |
+
│ . . C1 │ │ . C1 │ │ .C1 . │
|
| 231 |
+
│ . . . │ │ . . . │ │ . . . │
|
| 232 |
+
│ C2 │ │ │ │ │
|
| 233 |
+
│ . . . . │ │ . . C2. . │ │ . . C2. . │
|
| 234 |
+
│ . . │ │ . . │ │ . . │
|
| 235 |
+
└──────────────┘ └──────────────┘ └──────────────┘
|
| 236 |
+
C = Centroid, . = Data point
|
| 237 |
+
```
|
| 238 |
+
|
| 239 |
+
### K-Means++: Khởi tạo thông minh
|
| 240 |
+
|
| 241 |
+
```
|
| 242 |
+
K-Means++ (Arthur & Vassilvitskii, 2007):
|
| 243 |
+
|
| 244 |
+
1. Chọn centroid đầu tiên random
|
| 245 |
+
2. Cho mỗi centroid tiếp theo:
|
| 246 |
+
- Tính D(x) = khoảng cách từ x đến centroid gần nhất
|
| 247 |
+
- Chọn point mới với xác suất ∝ D(x)²
|
| 248 |
+
→ Centroids xa nhau → converge nhanh hơn
|
| 249 |
+
```
|
| 250 |
+
|
| 251 |
+
### Chọn K tối ưu
|
| 252 |
+
|
| 253 |
+
```
|
| 254 |
+
Elbow Method: Silhouette Method:
|
| 255 |
+
|
| 256 |
+
Inertia Silhouette Score
|
| 257 |
+
│ │
|
| 258 |
+
│\ │ *
|
| 259 |
+
│ \ │ * *
|
| 260 |
+
│ \ │ * *
|
| 261 |
+
│ \ │* *
|
| 262 |
+
│ \_____ │
|
| 263 |
+
│ ↑ │
|
| 264 |
+
│ Elbow │
|
| 265 |
+
│ \_____ │
|
| 266 |
+
└─────────────── K └─────────────── K
|
| 267 |
+
|
| 268 |
+
K* = point where adding K* = max silhouette
|
| 269 |
+
more clusters gives s(i) = (b(i)-a(i)) / max(a(i),b(i))
|
| 270 |
+
diminishing returns a(i) = avg distance to same cluster
|
| 271 |
+
b(i) = avg distance to nearest other cluster
|
| 272 |
+
```
|
| 273 |
+
|
| 274 |
+
---
|
| 275 |
+
|
| 276 |
+
## 5.6. DBSCAN (Density-Based Clustering)
|
| 277 |
+
|
| 278 |
+
### Nguyên lý hoạt động
|
| 279 |
+
|
| 280 |
+
```
|
| 281 |
+
DBSCAN Parameters:
|
| 282 |
+
- ε (epsilon): Bán kính neighborhood
|
| 283 |
+
- MinPts: Số điểm tối thiểu trong ε-neighborhood
|
| 284 |
+
|
| 285 |
+
3 loại điểm:
|
| 286 |
+
┌──────────────────────────────────────┐
|
| 287 |
+
│ │
|
| 288 |
+
│ Core Point (≥ MinPts trong ε) │
|
| 289 |
+
│ ┌─────┐ │
|
| 290 |
+
│ │ ●──┼──. . . (≥5 points) │
|
| 291 |
+
│ │ ε │ │
|
| 292 |
+
│ └─────┘ │
|
| 293 |
+
│ │
|
| 294 |
+
│ Border Point (< MinPts nhưng │
|
| 295 |
+
│ trong ε của Core Point) │
|
| 296 |
+
│ ○ ← border │
|
| 297 |
+
│ ↓ │
|
| 298 |
+
│ ┌─────┐ │
|
| 299 |
+
│ │ ● │ core │
|
| 300 |
+
│ └─────┘ │
|
| 301 |
+
│ │
|
| 302 |
+
│ Noise Point (không thuộc cluster) │
|
| 303 |
+
│ ✕ ← noise/outlier │
|
| 304 |
+
│ │
|
| 305 |
+
└──────────────────────────────────────┘
|
| 306 |
+
|
| 307 |
+
DBSCAN Algorithm:
|
| 308 |
+
1. Cho mỗi point chưa visited:
|
| 309 |
+
a. Đánh dấu visited
|
| 310 |
+
b. Tìm ε-neighborhood
|
| 311 |
+
c. Nếu ≥ MinPts → Core point → Tạo/mở rộng cluster
|
| 312 |
+
d. Nếu < MinPts → Tạm đánh dấu noise (có thể thành border sau)
|
| 313 |
+
|
| 314 |
+
DBSCAN vs K-Means:
|
| 315 |
+
┌────────────────┬─────────────┬──────────────┐
|
| 316 |
+
│ Feature │ K-Means │ DBSCAN │
|
| 317 |
+
├────────────────┼─────────────┼──────────────┤
|
| 318 |
+
│ # Clusters │ Phải chỉ K │ Tự phát hiện │
|
| 319 |
+
│ Cluster shape │ Spherical │ Arbitrary │
|
| 320 |
+
│ Outlier handle │ Không │ Có (noise) │
|
| 321 |
+
│ Scalability │ O(nKt) │ O(n²) worst │
|
| 322 |
+
│ Parameters │ K │ ε, MinPts │
|
| 323 |
+
└────────────────┴─────────────┴──────────────┘
|
| 324 |
+
```
|
| 325 |
+
|
| 326 |
+
---
|
| 327 |
+
|
| 328 |
+
## 5.7. Phân cụm Phân cấp (Hierarchical Clustering)
|
| 329 |
+
|
| 330 |
+
```
|
| 331 |
+
Agglomerative (Bottom-up):
|
| 332 |
+
|
| 333 |
+
Start: Mỗi point = 1 cluster
|
| 334 |
+
|
| 335 |
+
Step 1: {A}{B}{C}{D}{E} Merge closest: A,B
|
| 336 |
+
Step 2: {A,B}{C}{D}{E} Merge closest: D,E
|
| 337 |
+
Step 3: {A,B}{C}{D,E} Merge closest: C,{D,E}
|
| 338 |
+
Step 4: {A,B}{C,D,E} Merge closest: {A,B},{C,D,E}
|
| 339 |
+
Step 5: {A,B,C,D,E} Done!
|
| 340 |
+
|
| 341 |
+
Dendrogram:
|
| 342 |
+
─────────────────────────
|
| 343 |
+
│ ┌───────────┤
|
| 344 |
+
│ ┌─────┤ │
|
| 345 |
+
│ │ │ ┌──────┤
|
| 346 |
+
│ ┌──┤ │ │ ┌───┤
|
| 347 |
+
─ A B C D E
|
| 348 |
+
|
| 349 |
+
Linkage Methods:
|
| 350 |
+
- Single: min distance between clusters
|
| 351 |
+
- Complete: max distance
|
| 352 |
+
- Average: mean distance (UPGMA)
|
| 353 |
+
- Ward's: minimize variance increase
|
| 354 |
+
```
|
| 355 |
+
|
| 356 |
+
---
|
| 357 |
+
|
| 358 |
+
## 5.8. Outlier Detection
|
| 359 |
+
|
| 360 |
+
### Phương pháp phát hiện Outlier
|
| 361 |
+
|
| 362 |
+
| Method | Type | Mô tả |
|
| 363 |
+
|--------|------|-------|
|
| 364 |
+
| **Z-Score** | Statistical | \|z\| > 3 → outlier |
|
| 365 |
+
| **Modified Z-Score** | Statistical (robust) | Uses MAD instead of std |
|
| 366 |
+
| **IQR** | Statistical | < Q1-1.5IQR or > Q3+1.5IQR |
|
| 367 |
+
| **Isolation Forest** | ML-based | Random splits, outliers isolated faster |
|
| 368 |
+
| **LOF** | Distance-based | Local Outlier Factor, density-based |
|
| 369 |
+
| **DBSCAN noise** | Density-based | Points classified as noise |
|
| 370 |
+
| **Mahalanobis Distance** | Statistical | Accounts for correlation |
|
| 371 |
+
|
| 372 |
+
---
|
| 373 |
+
|
| 374 |
+
## 🔬 Labs
|
| 375 |
+
|
| 376 |
+
- [`lab-05-apriori.py`](lab-05-apriori.py) — Apriori & FP-Growth trên Olist
|
| 377 |
+
- [`lab-05-decision-tree.py`](lab-05-decision-tree.py) — Decision Tree Classification
|
| 378 |
+
- [`lab-05-naive-bayes.py`](lab-05-naive-bayes.py) — Naive Bayes Classification
|
| 379 |
+
- [`lab-05-kmeans.py`](lab-05-kmeans.py) — K-Means Customer Segmentation
|
| 380 |
+
- [`lab-05-dbscan.py`](lab-05-dbscan.py) — DBSCAN Anomaly Detection
|
| 381 |
+
|
| 382 |
+
---
|
| 383 |
+
|
| 384 |
+
## 📝 Câu hỏi ôn tập
|
| 385 |
+
|
| 386 |
+
1. Tính Support, Confidence và Lift cho rule {A} → {B} biết: N=100, count(A)=40, count(B)=50, count(A∪B)=25
|
| 387 |
+
2. Giải thích nguyên lý Apriori. Tại sao nó giúp giảm số lượng candidates?
|
| 388 |
+
3. So sánh Information Gain (ID3) v�� Gain Ratio (C4.5). Tại sao Gain Ratio tốt hơn?
|
| 389 |
+
4. K-Means có nhược điểm gì? K-Means++ cải thiện như thế nào?
|
| 390 |
+
5. DBSCAN phát hiện outlier như thế nào? So sánh với K-Means về handling outliers.
|
| 391 |
+
6. Cho dataset: [1,2,3,100,2,3,1,2,500]. Dùng IQR method, xác định outliers.
|