SVM

The Sequential Minimal Optimization algorithm (SMO) was proposed by John Platt in 1998, decomposing large-scale quadratic programming problems into minimal sub-problems involving only two variables.

SVM finds the optimal hyperplane by maximizing the classification margin, and its kernel trick can implicitly map data to a high-dimensional space to handle nonlinearly separable problems.

Traditional models such as SVM or decision trees are extremely sensitive to label noise, because their capacity and optimization path lack the preference for learning commonalities first and memorizing special cases later.

The complexity of computing the Gram matrix in traditional kernel methods (such as SVM using the RBF kernel) is O(n²) or even O(n³), making it difficult to scale to large-scale data.

The combination of EfficientNet-B3 and SVM achieved the strongest baseline performance, with an accuracy of 96.31% and a macro F1 score of 96.36%.

Classical machine learning methods (such as logistic regression, SVM, and Naive Bayes), combined with pre-computed text features such as TF-IDF or fixed word vectors, have extremely small parameter counts, with CPU inference latency as low as the millisecond level.

SVMs were the dominant method for classification tasks before the rise of deep learning, and still find stable application today in small-sample, high-dimensional structured data scenarios.

The classical model competitors in the comparison experiment include algorithms such as SVM, XGBoost, and logistic regression.

Important theoretical work on foundational algorithms such as SVM and AdaBoost was published in JMLR.

In SVM, the data points closest to the decision boundary are called support vectors and are the only samples that truly influence the position of the boundary; even if the remaining data points are removed, the boundary does not change.

Distance- or gradient-based algorithms (KNN, SVM, logistic regression, neural networks) are sensitive to feature scale and require feature scaling.

SVM benefited significantly from PCA dimensionality reduction, with classification accuracy noticeably improving after principal component analysis processing.

The research uses three machine learning algorithms for training and evaluation: K-Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Machine (SVM).

SVMs perform excellently on high-dimensional sparse data but have longer training times for large-scale data, while random forests natively support parallelization and are less sensitive to hyperparameters

Machine learning algorithms such as SVM and KNN are highly sensitive to the units of features.

该SWD特征变换实验在决策树上表现良好，但在逻辑回归、SVM、KNN等算法上效果糟糕

KNN、SVM的RBF核等依赖距离度量的算法及逻辑回归等依赖线性可分假设的算法对全局几何结构扭曲敏感，SWD变换会破坏其性能