|
| | Classifier | Features of classifier |
|
| 1 | RF | The effectiveness of random forests is due to several points that are as follows: Logging below a given node in a subset of the necessary features [35]. During the development of trees, the random forest contributes to the unpredictability of the model. When dividing a node, it finds the best property from a random set of attributes rather than looking for a key component. As a result, more variance leads to a better model [25, 26, 36]. It has powerful predictive performance with little processing and easy to understand. It is implemented using algorithms with built-in feature selection techniques [34]. The success of the random forest model is mainly because many non-correlated (trees) models outperform any single component models. Random forests are frequently used for trait selection in the data science process due to the low correlation between models [35]. |
| 2 | KNN | KNN is a supervised approach that predicts the output of data points using a labelled input dataset. It is one of the most basic machine learning algorithms, and it may be used for a wide range of issues. It is primarily based on visual resemblance. The training data is saved and only used to produce real-time predictions to learn from it. As a result, the KNN technique is substantially quicker than earlier training-based algorithms [32] |
| 3 | DTC | The Decision Tree Classifier’s key benefit is its ability to use a variety of feature subsets and decision rules at different categorization stages. A generic Decision Tree comprises one root node, several internal and leaf nodes, and branches [33] |
| 4 | SVM | SVM generates a decision boundary, or a hyperplane, between two classes to split or categorize them. SVM is also used in object identification and picture classification. The SVM method aims to find the optimal line or decision boundary for organizing n-dimensional space so that the following data points may be readily classified. A hyperplane is a boundary that is the optimum option. Why is the SVM classifier, in particular, the most effective classification approach for binary classification tasks? The best is determined by the data utilized and the circumstance; hence, no classifier can always be the best for all data and problems [34] |
| 5 | Mahalanobis | The Mahalanobis distance measures the degree of correlation between variables. You may have observed, for example, that gas mileage and displacement are significantly connected. As a result, the Euclidean distance computation contains much duplicate information [35] |
| Although the Mahalanobis-distance-based technique is motivated by classification prediction confidence, we discover that its improved performance is due to nonclassification information [36] |
|
| 6 | Normal Bayes | The Normal Bayes method is a classification approach for binary and multiclass data categorization. Naive Bayes outperforms numerical input variables when it comes to categorical input variables. It may be used to forecast data and make predictions based on historical data |
| Advantages. The class of the test dataset may be predicted quickly and easily. It is also good at predicting multiclass. A Naive Bayes classifier outperforms competing models such as logistic regression and takes less training data when the independence criteria are fulfilled [37] |
|
| 7 | MLC | The Maximum Likelihood Classifier is a popular remote sensing classification method that categorizes pixels with the highest likelihood into the proper class. The likelihood Lk is the posterior probability of a pixel belonging to class k [40] |
| 8 | Maximum Entropy | If only one parameter about a probability distribution is known, the principle of Maximum Entropy is a model generation criterion that involves picking the most unexpected (Maximum Entropy) prior assumption [41] |
| 9 | ANN | ANN learning approach teaches you how to achieve a complicated goal or optimize a given dimension over a long period. Nonparametric ANN techniques offer a particular advantage over statistical classification methods in that they do not require a priori knowledge of the distribution model of input data [42] |
| 10 | MDC | It gives classification with the fewest total parameters and computing demand but at the cost of accuracy. MDC’s purpose is to categorize as many patterns as possible properly. The MDC technique identifies centroids of classes and calculates distances between them and the test pattern [43] |
| 11 | SAM | It is a method for matching picture spectra to known spectra or an automated end member. The SAM classification yields a picture displaying the best match at each pixel [44] |
| 12 | Parallelepiped | It uses the class signature threshold to identify whether a pixel belongs to a class, is mixed up with other styles, or is unclassified [45] |
|
