What is Classification?

What Is Classification?

Classification is one of the two primary types of supervised machine learning (the other being regression). In a classification task, the model learns to assign inputs to predefined categories or classes. Given a set of features about a data point, the model predicts which category it belongs to.

Examples are everywhere in business:

• Email filtering -- spam or not spam
• Fraud detection -- fraudulent or legitimate
• Customer sentiment -- positive, neutral, or negative
• Lead qualification -- hot, warm, or cold
• Medical diagnosis -- condition present or absent

Types of Classification

Binary Classification

The simplest form: two possible classes.

• Fraud detection: fraudulent vs. legitimate
• Customer churn: will churn vs. will stay
• Credit approval: approve vs. reject
• Email: spam vs. not spam

Multi-Class Classification

More than two possible classes, with each input belonging to exactly one.

• Customer segmentation: premium, standard, basic, at-risk
• Document categorization: invoice, contract, receipt, correspondence
• Product categorization: assigning products to one of dozens or hundreds of categories

Multi-Label Classification

Each input can belong to multiple classes simultaneously.

• News articles tagged with multiple topics (politics, economy, technology)
• Products assigned multiple attribute tags (waterproof, lightweight, premium)
• Customer support tickets categorized by multiple issue types

Common Classification Algorithms

Different algorithms suit different scenarios:

• Logistic Regression -- Despite the name, this is a classification algorithm. Simple, fast, interpretable. Good baseline for binary classification.
• Decision Trees -- Intuitive, interpretable, handle mixed data types. Can overfit but form the basis for powerful ensemble methods.
• Random Forests -- Ensemble of many decision trees. Robust, accurate, handles noisy data well. One of the most reliable general-purpose classifiers.
• Gradient Boosting (XGBoost, LightGBM) -- Sequential ensemble that builds trees to correct previous errors. Often the top performer on tabular data.
• Support Vector Machines (SVM) -- Effective for high-dimensional data. Less commonly used today due to scalability limitations.
• Neural Networks -- Most powerful for complex, unstructured data (images, text). Require more data and compute than other approaches.

Evaluating Classification Models

Accuracy alone can be misleading. Consider a fraud detection system where only 1% of transactions are fraudulent. A model that predicts "not fraud" for every transaction achieves 99% accuracy but catches zero fraud.

Key evaluation metrics include:

• Precision -- Of all predictions of class X, what percentage was actually class X? High precision means fewer false alarms.
• Recall (Sensitivity) -- Of all actual class X examples, what percentage did the model catch? High recall means fewer missed detections.
• F1 Score -- The harmonic mean of precision and recall. Useful when you need to balance both.
• AUC-ROC -- Measures the model's ability to distinguish between classes across all threshold settings.
• Confusion Matrix -- A table showing exactly where the model gets things right and wrong for each class.

The right metric depends on the business cost of errors. In fraud detection, missing a fraudulent transaction (low recall) may cost more than flagging a legitimate one (low precision). In medical diagnosis, a false negative could be life-threatening.

Business Applications Across Southeast Asia

Classification powers high-impact applications throughout the ASEAN region:

• Credit scoring and risk assessment -- Banks and fintech companies in Indonesia, the Philippines, and Vietnam classify loan applicants into risk categories, enabling lending to underbanked populations. Companies like Akulaku, Kredivo, and Atome use classification models to assess risk for millions of users.
• Customer churn prediction -- Telecom providers and subscription businesses classify customers by churn risk, enabling proactive retention. In markets with intense competition like Thailand and Indonesia, reducing churn by even 5% significantly impacts revenue.
• Fraud detection -- Digital payment platforms (GrabPay, GoPay, ShopeePay, Dana) classify transactions in real time to catch fraud while minimizing friction for legitimate users.
• Content moderation -- Social platforms and marketplaces classify user-generated content to detect policy violations, fake listings, and harmful content across multiple languages.
• Document classification -- Enterprises classify incoming documents (contracts, invoices, permits) for automated routing and processing, especially valuable for organizations handling multilingual documents.

Building a Classification System: Practical Steps

1. Define the classes clearly -- Ambiguous class definitions lead to poor labeling and poor models
2. Collect and label training data -- Ensure sufficient examples of each class, especially minority classes
3. Handle class imbalance -- In many business problems, one class is much rarer than others (e.g., fraud). Techniques include oversampling, undersampling, SMOTE, and adjusting class weights.
4. Choose the right metric -- Align the evaluation metric with business costs of different types of errors
5. Set the classification threshold -- Most models output probabilities. Adjusting the threshold trades off precision and recall to match business requirements.
6. Monitor in production -- Class distributions can shift over time. Monitor performance and retrain as needed.

The Bottom Line

Classification is the most commercially deployed type of machine learning. If your business needs to make categorical decisions at scale -- approve or reject, flag or pass, categorize or route -- classification is your tool. Its mature ecosystem of algorithms, evaluation tools, and deployment patterns makes it an ideal starting point for organizations beginning their ML journey.