Algorithm Summary, analogy

1. Bayesian Classifier for document classification:

Monitoring algorithm Advantages: efficient training and querying data each training session may only be a training item, not the decision tree, the SVM must pass in the whole group, in order to get the final result disadvantage : Because Bayes theorem assumes that features are independent of each other, it is not possible to classify the results of combinatorial features 2. Decision Tree classifier: Supervisory algorithm: Advantages: The interpretation of the model is relatively easy, The most important factors are well positioned near the root, and it is very clear what variables are best used to split the data, and it is very intentional to be able to process both categorical and numerical data at the same time in advertising planning and judging which data should be collected.It is very easy to deal with the interaction between variables, and the Bayesian classification of the score file classification is good in this respect .Cons: When it comes to classifying attributes with a lot of data, it is difficult not to support incremental training 3. Neural network supervision, can identify important information and non-important information, can be used for classification, can also be used for numerical pre- The advantages of the problem: dealing with complex nonlinear functions, and the dependency relationship between different inputs incremental training disadvantage: black-box method, Controllability 4. Support Vector machines, nuclear techniques: Advantages: The fastest classification of new observational dataClassification for large data sets, decision trees and other classification methods better suited for small datasetsDisadvantage: For different datasets It is necessary to re-determine these (kernel) transformation functions and their parameters is also the black box technology 5.kNN advantages: using complex functions into                    While maintaining a simple and easy-to-understand feature, you can know the importance of each attribute online technical disadvantage: requires a lot of training data Finding a scaling factor for large datasets is computationally significant 6. Clustering 7. Non-negative matrix factorization: Splitting the data to get a new relationship 8. Optimization

Algorithm Summary, analogy