Machine Learning Fundamental 3

Pertemuan 3: Data Preprocessing

🎯 Tujuan Pembelajaran

Setelah mengikuti pertemuan ini, mahasiswa mampu:

1. Memahami pentingnya preprocessing dalam pipeline machine learning
2. Menangani missing value secara tepat
3. Melakukan encoding data kategorikal
4. Melakukan feature scaling
5. Melakukan train-test split dengan benar
6. Memahami potensi data leakage

---

🧠 1. Mengapa Data Preprocessing Penting?

Dalam praktik nyata, data hampir tidak pernah bersih.
Masalah umum:

• Missing value
• Outlier
• Data kategorikal
• Skala fitur berbeda
• Data tidak seimbang

Jika preprocessing salah → model:

• Overfitting
• Underfitting
• Bias
• Tidak stabil

Pipeline umum ML:

EDA → Preprocessing → Feature Engineering → Modeling → Evaluation

Preprocessing adalah fondasi sebelum modeling.

---

📊 2. Menangani Missing Value

🔎 Mengecek Missing Value

import pandas as pd
import seaborn as sns

df = sns.load_dataset("titanic")

df.isnull().sum()

---

🛠 Strategi Penanganan Missing Value

1️⃣ Menghapus Data

df.dropna(inplace=True)

Digunakan jika:

• Missing sangat sedikit
• Tidak memengaruhi distribusi data

---

2️⃣ Imputasi Mean / Median

df["age"].fillna(df["age"].median(), inplace=True)

Digunakan untuk data numerik.

• Mean → jika distribusi normal
• Median → jika ada outlier

---

3️⃣ Imputasi Modus (Kategorikal)

df["embarked"].fillna(df["embarked"].mode()[0], inplace=True)

---

🔤 3. Encoding Data Kategorikal

Machine learning hanya menerima angka.

🎯 Label Encoding

Digunakan untuk kategori biner.

from sklearn.preprocessing import LabelEncoder

le = LabelEncoder()
df["sex"] = le.fit_transform(df["sex"])

Output:

• male → 1
• female → 0

---

🎯 One Hot Encoding

Digunakan untuk kategori > 2.

df = pd.get_dummies(df, columns=["embarked"], drop_first=True)

Kenapa drop_first=True? → Menghindari dummy variable trap (multicollinearity).

---

📏 4. Feature Scaling

Beberapa algoritma sensitif terhadap skala fitur:

• KNN
• SVM
• Logistic Regression
• Neural Network

Tidak sensitif:

• Decision Tree
• Random Forest

---

📌 Standardization (Z-score)

from sklearn.preprocessing import StandardScaler

scaler = StandardScaler()

numerical_cols = ["age", "fare"]
df[numerical_cols] = scaler.fit_transform(df[numerical_cols])

Formula:
z = \frac{x - \mu}{\sigma}

---

📌 Normalization (MinMaxScaler)

from sklearn.preprocessing import MinMaxScaler

scaler = MinMaxScaler()
df[numerical_cols] = scaler.fit_transform(df[numerical_cols])

Range: 0 – 1

---

✂️ 5. Train-Test Split

Tujuan:

• Menghindari overfitting
• Mengukur performa model pada data baru

from sklearn.model_selection import train_test_split

X = df.drop("survived", axis=1)
y = df["survived"]

X_train, X_test, y_train, y_test = train_test_split(
    X, y,
    test_size=0.2,
    random_state=42
)

Artinya:

• 80% training
• 20% testing

---

⚠️ 6. Data Leakage (Konsep Penting!)

Kesalahan umum:
❌ Scaling sebelum train-test split
❌ Menggunakan seluruh data untuk imputasi

Yang benar:
Split dulu → Fit hanya di data training → Transform training & test

Contoh benar:

scaler = StandardScaler()

X_train[numerical_cols] = scaler.fit_transform(X_train[numerical_cols])
X_test[numerical_cols] = scaler.transform(X_test[numerical_cols])

---

🧪 Praktikum Lengkap (Ringkas)

import pandas as pd
import seaborn as sns
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler, LabelEncoder

df = sns.load_dataset("titanic")

# Handle missing
df["age"].fillna(df["age"].median(), inplace=True)
df["embarked"].fillna(df["embarked"].mode()[0], inplace=True)

# Encoding
le = LabelEncoder()
df["sex"] = le.fit_transform(df["sex"])
df = pd.get_dummies(df, columns=["embarked"], drop_first=True)

# Split
X = df.drop("survived", axis=1)
y = df["survived"]

X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42
)

# Scaling
scaler = StandardScaler()
numerical_cols = ["age", "fare"]

X_train[numerical_cols] = scaler.fit_transform(X_train[numerical_cols])
X_test[numerical_cols] = scaler.transform(X_test[numerical_cols])

---

📝 Tugas Mandiri

1. Lakukan preprocessing lengkap pada dataset apapun di Kaggle.
2. Jelaskan:
   • Mengapa encoding diperlukan?
   • Mengapa scaling diperlukan?
   • Apa itu data leakage?
3. Dokumentasikan dalam template: https://github.com/AzharRizkiZ/Template-DS-ML
4. Upload ke GitHub.

---

🎓 Target Kompetensi Setelah Pertemuan 3

Siswa mampu:

• Menyiapkan dataset real untuk modeling
• Menghindari kesalahan preprocessing
• Memahami pipeline machine learning dasar