Classification and Prediction

Classification vs. Prediction

n  Classification: 

n  predicts categorical class labels

n  classifies data (constructs a model) based on the training set and the values (class labels) in a classifying attribute and uses it in classifying new data

n  Prediction: 

n  models continuous-valued functions, i.e., predicts unknown or missing values

n  Typical Applications

n  credit approval

n  target marketing

n  medical diagnosis

n  treatment effectiveness analysis

Classification—A Two-Step Process

n  Model construction: describing a set of predetermined classes

n  Each tuple/sample is assumed to belong to a predefined class, as determined by the class label attribute

n  The set of tuples used for model construction: training set

n  The model is represented as classification rules, decision trees, or mathematical formulae

n  Model usage: for classifying future or unknown objects

n  Estimate accuracy of the model

n  The known label of test sample is compared with the classified result from the model

n  Accuracy rate is the percentage of test set samples that are correctly classified by the model

n  Test set is independent of training set, otherwise over-fitting will occur

Classification Process (): Model Construction

Classification Process (): Use the Model in Prediction

Supervised vs. Unsupervised Learning

n  Supervised learning (classification)

n  Supervision: The training data (observations, measurements, etc.) are accompanied by labels indicating the class of the observations

n  New data is classified based on the training set

n  Unsupervised learning (clustering)

n  The class labels of training data is unknown

n  Given a set of measurements, observations, etc. with the aim of establishing the existence of classes or clusters in the data

Issues regarding classification and prediction (1): Data Preparation

n  Data cleaning

n  Preprocess data in order to reduce noise and handle missing values

n  Relevance analysis (feature selection)

n  Remove the irrelevant or redundant attributes

n  Data transformation

n  Generalize and/or normalize data

Issues regarding classification and prediction (2): Evaluating Classification Methods

n  Predictive accuracy

n  Speed and scalability

n  time to construct the model

n  time to use the model

n  Robustness

n  handling noise and missing values

n  Scalability

n  efficiency in disk-resident databases

n  Interpretability:

n  understanding and insight provded by the model

n  Goodness of rules

n  decision tree size

n  compactness of classification rules