Scikit-learn Preprocessing Techniques for ML ⚡

1. Standardization with StandardScaler
Standardization transforms data so that it has mean = 0 and standard deviation = 1, which helps many ML algorithms converge faster.

# Long form
from sklearn.preprocessing import StandardScaler
import numpy as np
data = np.array([[1,2],[3,4],[5,6]])
scaler = StandardScaler()
scaled = scaler.fit_transform(data)

# One-liner
scaled = __import__('sklearn.preprocessing').StandardScaler().fit_transform([[1,2],[3,4],[5,6]])


This ensures features are on the same scale, improving model performance.

2. Normalization with MinMaxScaler
Normalization scales features to a specific range (usually [0,1]), often required for neural networks.

# Long form
from sklearn.preprocessing import MinMaxScaler
data = [[1,2],[3,4],[5,6]]
scaler = MinMaxScaler()
normalized = scaler.fit_transform(data)

# One-liner
normalized = __import__('sklearn.preprocessing').MinMaxScaler().fit_transform([[1,2],[3,4],[5,6]])


It ensures all features contribute equally to model training.

3. Encoding categorical data
Machine learning models can’t handle strings, so categorical values must be encoded.

# Long form
from sklearn.preprocessing import LabelEncoder, OneHotEncoder
import numpy as np
labels = ['cat','dog','cat','bird']
label_enc = LabelEncoder().fit_transform(labels)
onehot_enc = OneHotEncoder(sparse=False).fit_transform(np.array(labels).reshape(-1,1))

# One-liner
le,oh = __import__('sklearn.preprocessing').LabelEncoder().fit_transform(['cat','dog','cat','bird']), __import__('sklearn.preprocessing').OneHotEncoder(sparse=False).fit_transform(np.array(['cat','dog','cat','bird']).reshape(-1,1))


Label encoding converts categories into numbers, while one-hot encoding creates binary vectors.

4. Handling missing values with SimpleImputer
Missing data is common in real-world datasets, and imputers can replace them with mean, median, or a constant.

# Long form
from sklearn.impute import SimpleImputer
import numpy as np
data = [[1,2],[np.nan,3],[7,6]]
imputer = SimpleImputer(strategy='mean')
imputed = imputer.fit_transform(data)

# One-liner
imputed = __import__('sklearn.impute').SimpleImputer(strategy='mean').fit_transform([[1,2],[None,3],[7,6]])


This prevents models from failing due to NaN values.

5. Splitting dataset with train_test_split
Splitting datasets into training and testing sets ensures unbiased evaluation of models.

# Long form
from sklearn.model_selection import train_test_split
X = [[1],[2],[3],[4],[5]]
y = [0,0,1,1,1]
X_train, X_test, y_train, y_test = train_test_split(X,y,test_size=0.2,random_state=42)

# One-liner
X_train,X_test,y_train,y_test = __import__('sklearn.model_selection').train_test_split([[1],[2],[3],[4],[5]],[0,0,1,1,1],test_size=0.2,random_state=42)


This is crucial for avoiding overfitting and ensuring reliable performance metrics.

5 Interview Questions (with Answers):

Q: What’s the difference between standardization and normalization?
A: Standardization → mean=0, std=1. Normalization → rescaling features to a fixed range (e.g., [0,1]).

Q: When would you use OneHotEncoder over LabelEncoder?
A: Use OneHotEncoder when categories don’t have an inherent order (e.g., colors), LabelEncoder when order matters.

Q: How do you handle missing values in scikit-learn?
A: Using SimpleImputer with strategies like mean, median, or most frequent.

Q: Why do we split data into training and testing sets?
A: To evaluate model performance on unseen data and prevent overfitting.

Q: Why is scaling important in machine learning?
A: Many algorithms (e.g., KNN, SVM, Gradient Descent) are sensitive to feature scales, so scaling ensures fair contribution.

#scikitlearn #Python #MachineLearning #DataScience #MLTips #CodingTips #InterviewPrep