News classification
Contact us
- Add: No. 9, North Fourth Ring Road, Haidian District, Beijing. It mainly includes face recognition, living detection, ID card recognition, bank card recognition, business card recognition, license plate recognition, OCR recognition, and intelligent recognition technology.
- Tel: 13146317170 廖经理
- Fax:
- Email: 398017534@qq.com
Several machine learning classification models
Several machine learning classification models
KNN: Dependent on data, no mathematical model at all. Suitable for models that are easy to interpret.
Sensitive to outliers, vulnerable to data imbalances.
Bayesian: Based on conditional probability, it is suitable for time divisions with small correlations between different dimensions, which is easier to explain. It is also suitable for incremental exercise, no need to recalculate it again. Application: Dregs mail disposal.
Decision Tree: This model makes it easier to understand the level of impact of different attributes on the results (as in the first few layers). Ability to handle different types of data simultaneously. However, since the tracking result only needs to change the attributes of the leaf nodes, it is vulnerable to attack. Application: The cornerstone of other algorithms.
Random Forest: Random forests are random integrations of decision trees that improve their vulnerability to attack at a certain level. Applicable to time points where the data dimension is not too high (tens of tens) and you want to reach higher accuracy. There is no need to adjust too many parameters, it is appropriate to use the time before you know what to apply.
SVM: SVM tries to stick to the spacing between samples as much as possible, and it is strong against attacks. Like RandomForest, it is the first way to try.
Logistic regression: Logistic regression, not only can output results but also output its corresponding probability. The fitted parameters clearly show the effect of each feature on the results. But it is essentially a linear classifier, which is not applicable when the correlation between features is high. Also pay attention to the impact of outliers.
Discriminat Analysis is typically LDA, which projects high-dimensional data onto low-dimensional data to make the data as close as possible. Often used as a dimensionality reduction tool. But keep in mind that LDA assumes that the data is normally spread.
Neural Network. It is also a black box, which is suitable for time-sharing applications with large data volume.
Ensemble-Boosting: Every time you find a classifier that can handle the current error, and then add weights. The benefit is that it comes with feature selection and discovers effective features. Also take a look at high-dimensional data.
Ensemble-Bagging: Exercise multiple weak classifiers to vote. Randomly select exercise sets to prevent overfitting.
Ensemble-Stacking: Take the results of the classifier as input and exercise a classifier. The usual last layer uses logistic Regression. It is possible to make a transition fit and rarely use it.
Sensitive to outliers, vulnerable to data imbalances.
Bayesian: Based on conditional probability, it is suitable for time divisions with small correlations between different dimensions, which is easier to explain. It is also suitable for incremental exercise, no need to recalculate it again. Application: Dregs mail disposal.
Decision Tree: This model makes it easier to understand the level of impact of different attributes on the results (as in the first few layers). Ability to handle different types of data simultaneously. However, since the tracking result only needs to change the attributes of the leaf nodes, it is vulnerable to attack. Application: The cornerstone of other algorithms.
Random Forest: Random forests are random integrations of decision trees that improve their vulnerability to attack at a certain level. Applicable to time points where the data dimension is not too high (tens of tens) and you want to reach higher accuracy. There is no need to adjust too many parameters, it is appropriate to use the time before you know what to apply.
SVM: SVM tries to stick to the spacing between samples as much as possible, and it is strong against attacks. Like RandomForest, it is the first way to try.
Logistic regression: Logistic regression, not only can output results but also output its corresponding probability. The fitted parameters clearly show the effect of each feature on the results. But it is essentially a linear classifier, which is not applicable when the correlation between features is high. Also pay attention to the impact of outliers.
Discriminat Analysis is typically LDA, which projects high-dimensional data onto low-dimensional data to make the data as close as possible. Often used as a dimensionality reduction tool. But keep in mind that LDA assumes that the data is normally spread.
Neural Network. It is also a black box, which is suitable for time-sharing applications with large data volume.
Ensemble-Boosting: Every time you find a classifier that can handle the current error, and then add weights. The benefit is that it comes with feature selection and discovers effective features. Also take a look at high-dimensional data.
Ensemble-Bagging: Exercise multiple weak classifiers to vote. Randomly select exercise sets to prevent overfitting.
Ensemble-Stacking: Take the results of the classifier as input and exercise a classifier. The usual last layer uses logistic Regression. It is possible to make a transition fit and rarely use it.