Building a Two-Dimensional SVM Classifier with Newton Interpolation

Summary

The article describes a compact support vector machine classification approach for two-dimensional trading data, implemented in MQL5 without an external library. It outlines polynomial kernels and support vectors, then uses Newton polynomial interpolation to construct a decision boundary between labeled data sets. The resulting boundary is applied to price features based on changes in high and low values to generate expert signal classes. Two extensions add graded regression outputs based on proximity to known points and refine the boundary using a smaller set of points near the opposing class.

The article reports a comparison of three test approaches, stating that the support-vector version has nearly ten times the expectancy of the other two. It does not provide enough detail here to establish the robustness of that result or its out-of-sample reliability. The author also notes that higher polynomial degrees and high-dimensional inputs can overfit, and that this two-dimensional demonstration does not cover the more complex radial basis function kernel commonly used in SVM applications.

Key ideas

• The method uses Newton polynomial interpolation to estimate a separating boundary for two-dimensional labeled data.
• Polynomial kernels map relationships between points, while support vectors identify boundary points near the other class.
• The example uses changes in high and low prices as its two input features.
• Regression-style outputs can reflect proximity to known vectors instead of returning only hard class labels.
• The reported expectancy comparison is specific to the article's tests and does not establish out-of-sample performance.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.