Loading
MindXchange
MindXchange
MindXchange

Quipoin Menu

Learn • Practice • Grow

Tutorial Exercises MCQs Interview Q & A Tests
machine-learning / Handling Categorical Data
tutorial
VideosDiscussion

Handling Categorical Data

Machine learning models understand only numbers. Text categories like "red", "blue", "green" or "low", "medium", "high" must be converted to numeric form. This chapter covers the two main techniques: label encoding and one‑hot encoding.

Label Encoding

Assign each unique category an integer (0,1,2,...).
from sklearn.preprocessing import LabelEncoder

le = LabelEncoder()
df['color_encoded'] = le.fit_transform(df['color'])
# 'red' → 0, 'blue' → 1, 'green' → 2
Use when categories have an order (ordinal), e.g., "low" < "medium" < "high". Avoid for nominal categories (no order) because models might assume 0 < 1 < 2.

One‑Hot Encoding

Create a new binary column for each category. Each column is 1 if the category is present, else 0.
# Using pandas get_dummies
df_encoded = pd.get_dummies(df, columns=['color'])

# Using scikit-learn
from sklearn.preprocessing import OneHotEncoder
encoder = OneHotEncoder(sparse_output=False)
encoded = encoder.fit_transform(df[['color']])
Best for nominal categories without order. Creates multiple columns – may increase dataset size.

When to Use Which?

ScenarioRecommended
Ordinal categories (small, medium, large)Label Encoding
Nominal categories (colors, countries)One‑Hot Encoding
High cardinality (many unique values)Frequency encoding or embeddings

Example: Predicting House Prices

Neighborhood (nominal) → one‑hot encode. Quality rating (poor, average, good) → label encode.


Two Minute Drill
  • Categorical data must be numeric for ML models.
  • Label encoding assigns integers (for ordinal data).
  • One‑hot encoding creates binary columns (for nominal data).
  • Use pd.get_dummies() or OneHotEncoder.

Need more clarification?

Drop us an email at career@quipoinfotech.com