Support Vector Machine

Repository from study in SVM techniques, with the objective to classsify data from Eletrocardiograms.

The code implemented here uses an SMO algorithm to resolve the optimization problems to approximate the Lagrange Multipliers to solve the classification problem.

Algorithms Types

• SMO: Sequential Minimal Optimization can be used when is not expected outliers in the data.
• ISDA: Iterative Single Data Algorithm can set a fraction of expected outliers in the data
• quadprog: Quadratic Programming can be use to increase the degree of precision

Recommended to use a optmized library (libsvm).

Kernels

The kernels can be a solution when the data don't have a explicit linear separation between the classes. They include functions like:

• Polynomials G(x₁,x₂) = (1 + x₁′x₂)p.

• Radial basis (Gaussian): G(x₁,x₂) = exp(–∥x₁–x₂)∥²).

• Multilayer perceptron or sigmoid (neural network) G(x₁,x₂) = tanh(p₁x₁′x₂ + p₂).

---

Folders Structure

This repository have the follows folders:

• _articles - reference studies on all the topics used in this research:

  • Correntropy
  • SVM
  • Wavelet

• _code - Research code fucntions

• _data - with the database data and files of post-processing

  • Arritmia -
  • Normal -

• _samples - example algorithms to work with the database

• _svmImplementation - SVM implementation code using a SMO algorithm