Class Prediction Task

A Class Prediction Task is a nominal prediction task whose mapping function assigns input items to class members from a finite class set.

• AKA: Classification Task, Categorization Task, Categorical Label Assignment Task, Class Assignment Task, Categorical Prediction Task.
• Context:
  • Input: an item (often a data item); and a class set.
    • optional input: a classification function structure; or a training set.
  • output: one or more class set members (possibly as class member identifiers).
    • optional output: a classification function structure.
  • measures: e.g. Classification Accuracy.
    • optional measures: Precision, Recall, F1 Score, ROC Curve for binary classification.
  • It can range from being a Binary Classification Task to being a Multiclass Classification Task (such as large multiclass classification), depending on the class set size.
  • It can range from being a Unilabel Classification Task to being a Multilabel Classification Task, depending on the number of class outputs.
  • It can range from being a Simple-Input Classification Task to being a Complex-Input Classification Task, depending on the complexity of the input item.
  • It can range from being a Model-based Classification Task to being a Rule-based Classification Task, depending on its decision mechanism.
  • It can range from being a Manual Classification Task to being an Automated Classification Task, depending on its execution method.
  • It can range from being a Heuristic Classification Task to being a Data-Driven Classification Task (such as a supervised classification task), depending on whether some uncertainty is allowed in the decision.
  • It can range from being a Probabilistic Classification Task to being a Deterministic Classification Task, depending on its output type.
  • It can range from being a Batch Classification Task to being an Online Classification Task, depending on its processing mode.
  • It can range from being a Flat Classification Task to being a Hierarchical Classification Task, depending on its class structure.
  • It can range from being a Balanced Classification Task to being an Imbalanced Classification Task, depending on its class distribution.
  • It can typically be solved by a Classification System (that implements a classification algorithm).
  • It can be instantiated as a Class Prediction Act.
  • It can often be a Data Classification Task.
  • It can require feature extraction from raw input data.
  • It can utilize decision boundarys for class separation.
  • It can employ ensemble methods for prediction improvement.
  • It can incorporate cost-sensitive learning when misclassification costs vary.
  • It can support explainable classification through interpretable models.
  • It can enable transfer learning across related domains.
  • It can facilitate active learning through informative sample selection.
  • ...
• Example(s):
  • Text Classification Tasks, such as:
    • Spam Email Classification Task distinguishing spam from legitimate email.
    • Sentiment Analysis Task categorizing text as positive/negative/neutral.
    • Document Topic Classification Task assigning documents to predefined topics.
    • Named Entity Recognition Task classifying text spans as entity types.
  • Image Classification Tasks, such as:
    • Object Recognition Task identifying objects in images.
    • Medical Image Classification Task detecting diseases in medical scans.
    • Face Recognition Task identifying individuals from facial images.
    • Scene Classification Task categorizing image environments.
  • Audio Classification Tasks, such as:
    • Speech Recognition Task converting speech to text categories.
    • Music Genre Classification Task categorizing music by style.
    • Sound Event Detection Task identifying audio events.
  • Complex Input Classification Tasks, such as:
    • Part-of-Speech Tagging Task assigning grammatical categories.
    • Text Chunking Task identifying phrase boundaries.
    • Protein Function Prediction Task classifying protein roles.
  • Rules-based Classification Tasks, such as:
    • Given the Classification Model IF X is human THEN X is mortal. what is the class membership of X=Socrates?
    • Expert System Classification Task using knowledge bases.
    • Decision Tree Classification Task following rule paths.
  • Domain-Specific Classification Tasks, such as:
    • Biological Species Classification in taxonomy.
    • Credit Risk Classification Task in finance.
    • Customer Churn Prediction Task in business.
    • Disease Diagnosis Task in healthcare.
  • Time Series Classification Tasks, such as:
    • Activity Recognition Task from sensor data.
    • Anomaly Detection Task in system monitoring.
    • Pattern Recognition Task in financial markets.
  • Given the Patient Record p, classify whether the patient should or should not be given aspirin.
  • Guess the outcome of a Card Draw Experiment based on your intuition.
  • Guess the outcome of a Card Draw Experiment based on data of the card's past performance.
  • Guess whether Customer [math]\displaystyle{ X }[/math] will not pay their current bill by its due date, based on your intuition.
  • Guess whether Customer [math]\displaystyle{ X }[/math] will not pay their current bill by its due date based on data of the customer and other customer's past performance.
  • ...
• Counter-Example(s):
  • Ordinal Prediction Task, which predicts ordered categories rather than nominal classes.
  • Numeric Prediction Task, which outputs continuous values rather than discrete classes.
  • Regression Task, which estimates quantities rather than categories.
  • Item Scoring Task, which assigns continuous scores rather than class labels.
  • Point Estimation Task, which estimates parameters rather than classifying.
  • Complex Output Decisioning Task, which produces structured outputs beyond class labels.
  • Clustering Task, which discovers groups without predefined class labels.
  • Ranking Task, which orders items rather than classifying them.
  • Dimensionality Reduction Task, which transforms data without classification.
  • Feature Selection Task, which identifies relevant features without predicting classes.
• See: Class, Classification Algorithm, Classifier, Cost-Benefit Matrix, Loss Function, Scientific Classification, Confusion Matrix, Classification Performance Metric, Machine Learning Task, Supervised Learning Task.

---

---

References

2011

• (Drummond, 2011c) ⇒ Chris Drummond. (2011). “Classification.” In: (Sammut & Webb, 2011) p.168

2009

• (Wikipedia, 2009) ⇒ http://en.wikipedia.org/wiki/Classification_(biology)
  • Biological classification or scientific classification in biology, is a method by which biologists group and categorize species of organisms. Biological classification is a form of scientific taxonomy, but should be distinguished from folk taxonomy, which lacks scientific basis. …

2009

• (Wikipedia, 2009) ⇒ http://en.wikipedia.org/wiki/Classification_(literature)
  • Classification is a figure of speech linking a proper noun to a common noun using the or other articles.

2009

• http://wordnet.princeton.edu/perl/webwn
  • categorization: the act of distributing things into classes or categories of the same type
  • a group of people or things arranged by class or category
  • restriction imposed by the government on documents or weapons that are available only to certain authorized people

2009

• http://en.wiktionary.org/wiki/classification
  • The act of forming into a class or classes; a distribution into groups, as classes, orders, families, etc. …

2009

• http://wordnet.princeton.edu/perl/webwn
  • arrange or order by classes or categories; "How would you classify these pottery shards--are they prehistoric?"
  • declare unavailable, as for security reasons; "Classify these documents"
  • relegate: assign to a class or kind; "How should algae be classified?"; "People argue about how to relegate certain mushrooms"

2009

• http://en.wiktionary.org/wiki/classify
  • to identify by or divide into classes; to categorize; to declare something a secret, especially a government secret

2009

• (Hand, 2009) ⇒ David J. Hand. (2009). “Mismatched Models, Wrong Results, and Dreadful Decisions: On Choosing Appropriate Data Mining Tools.” In: Proceedings of the 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2009). doi:10.1145/1557019.1557021
  • QUOTE: For predictive classification problems, a wide variety of score functions exist, including measures such as precision and recall, the F measure, misclassification rate, the area under the ROC curve (the AUC), and others.

1986

• (Quinlan, 1986a) ⇒ J. Ross Quinlan. (1986). “Induction of Tecision Trees." Machine Learning, 1(1).
  • QUOTE: … Examples of classification tasks are: Page 3. … It might appear that classification tasks are only a minuscule subset of procedural tasks, but even activities such as robot planning can be recast as classification problems (Dechter and Michie, 1985). …

---