Advantages and disadvantages of machine learning algorithm and its application fields

Original: http://blog.csdn.net/mach_learn/article/details/39501849

Decision tree One, decision tree advantages

1, decision tree easy to understand and explain, can be visualized analysis, easy to extract rules.

2. Both nominal and numerical data can be processed at the same time.

3, test data set, run faster.

4, decision tree can be well extended to large database, and its size is independent of the database size.

Second, the decision tree disadvantage

1, to the missing data processing is difficult.

2, prone to the problem of fitting.

3. Ignore the correlation of attributes in the dataset.

4, the ID3 algorithm calculates the information gain when the result bias value is more characteristic.

Iii. measures of improvement

1, pruning the decision tree. Cross-validation and regularization methods can be used.

2, using combination algorithm based on decision tree, such as bagging algorithm, randomforest algorithm, can solve the problem of overfitting.

Iii. Fields of Application

Business management practice, enterprise investment decision-making, because the decision-making tree very good analytical ability, in the decision-making process application more.

KNN algorithm and the advantages of KNN algorithm

1, KNN is an online technology, new data can be directly added to the data set without retraining

2, KNN theory is simple, easy to achieve

The disadvantage of KNN algorithm

1, for large sample capacity of the data set calculation is relatively large.

2, when the sample is unbalanced, the prediction deviation is relatively large. such as: A certain class of samples are relatively small, and other types of samples are more.

3, KNN every time the classification will be re-global operation.

4, the choice of K value size.

Three, KNN algorithm application field

Text classification, pattern recognition, cluster analysis, multi-classification field

Support Vector Machine (SVM) one, SVM advantages

1. Solve the problem of machine learning under small sample.

2. Solve non-linear problems.

3, no local minimum problem. (relative to algorithms such as neural networks)

4, can be very good processing high-dimensional data sets.

5, the generalization ability is stronger than.

Second, the SVM disadvantage

1, the high-dimensional mapping of the kernel function is not strong, especially the radial basis function.

2, sensitive to missing data.

Iii. Application Fields of SVM

Text classification, image recognition, major two classification areas

AdaBoost algorithm one, AdaBoost algorithm advantages

1, a good use of the weak classifier cascade.

2, different classification algorithms can be used as weak classifiers.

3, AdaBoost has a high precision.

4, relative to the bagging algorithm and the random forest algorithm, adaboost fully consider the weight of each classifier.

Ii. Disadvantages of AdaBoost algorithm

1, the number of adaboost iterations is not very good set of weak classifiers, you can use cross-validation to determine.

2, the data imbalance causes the classification precision to descend.

3, the training is time-consuming, each time re-select the current classifier best segmentation point.

Three, adaboost application field

Pattern recognition, computer vision, for two classification and multi-classification scenarios

Naive Bayesian algorithm One, naive Bayesian algorithm advantages

1, for the large number of training and query with a higher speed. Even with very large training sets, there are usually only a relatively small number of features for each project, and the training and classification of the project is only a mathematical operation of the characteristic probabilities.

2, support incremental operation. The new samples can be trained in real time.

3. Naive Bayes is easy to understand the result explanation.

Second, naive Bayesian shortcomings

1. Due to the assumption that the sample attribute independence is used, it is not effective if the sample properties are associated.

Third, naive Bayesian application field

More use in text categorization and fraud detection

Logistic regression algorithm one, logistic regression advantage

1, the calculation cost is not high, easy to understand and realize

Second, logistic regression disadvantage

1, easy to produce under-fitting.

2, the classification accuracy is not high.

Third, logistic regression application field

For the two classification field, the probability value can be obtained, which is applicable to the field ranked according to the classification probability, such as search rankings.

The extended softmax of logistic regression can be applied to multi-classification fields, such as handwritten handwriting recognition and so on.

Artificial neural network One, the advantage of neural network

1, high classification accuracy, learning ability is very strong.

2, the noise data robustness and fault tolerance is strong.

3, has the associative ability, can approximate any nonlinear relation.

Ii. Disadvantages of neural networks

1, the neural network parameters, weight and threshold value.

2, black box process, can not observe the intermediate results.

3, the learning process is relatively long, it is possible to fall into the local minimum value.

Three, artificial Neural network application field

At present, deep neural network has been applied to computer vision, natural language processing, speech recognition and other fields and achieved good results.

Advantages and disadvantages of machine learning algorithm and its application fields