🧠 ml_svm – Native Dart Support Vector Machine (SVM)

A fully Dart-native implementation of the Support Vector Machine (SVM) algorithm.
This library supports binary and multi-class classification using hinge loss + L2 regularization, optimized with Float64List vectors and zero dependencies.

⚑ No Python, no bindings β€” just pure Dart for real machine learning on Flutter, CLI, or embedded apps.

πŸ“˜ What is SVM?

Support Vector Machine (SVM) is a supervised learning algorithm used for classification and regression.
It works by finding the hyperplane that best separates different classes in a high-dimensional space.

This package implements:

  • Linear SVM
  • Hinge Loss
  • L2 Regularization
  • Gradient Descent Optimization
  • One-vs-Rest (OvR) for multi-class classification

πŸš€ Features

βœ… Linear classification
βœ… Binary and multi-class (OvR) support
βœ… Optimized for Dart's Float64List
βœ… Gradient descent training
βœ… Intercept/bias term
βœ… Utility functions: normalization, shuffling
βœ… MIT License – Free for commercial or academic use

πŸ“¦ Installation

dart pub add ml_svm

⚑ Quick Example

import 'dart:typed_data';
import 'package:ml_svm/ml_svm.dart';

void main() {
  // Training dataset: [height, weight]
  final features = <Float64List>[
    Float64List.fromList([160, 50]),
    Float64List.fromList([170, 65]),
    Float64List.fromList([180, 80]),
    Float64List.fromList([155, 48]),
  ];

  final labels = ['underweight', 'normal', 'overweight', 'underweight'];

  // Normalize input features
  final normalized = normalize(features);

  // Train the SVM model
  final model = SVMModel(
    learningRate: 0.01,
    regularizationC: 1.0,
    maxIterations: 500,
  );

  model.fit(normalized, labels);

  // Predict a new sample
  final testSample = normalize([
    Float64List.fromList([172, 70])
  ]).first;

  print('Prediction: ${model.predict(testSample)}');
  print('Scores: ${model.decisionFunction(testSample)}');
}

πŸ“Œ Expected Output:

Prediction: normal
Scores: {underweight: -0.12, normal: 0.53, overweight: 0.42}

πŸ“š API Documentation

SVMModel Constructor

SVMModel({
  double learningRate = 0.01,
  double regularizationC = 1.0,
  int maxIterations = 1000,
  double tolerance = 1e-4,
});

Main Methods

Method Description
fit(List<Float64List> X, List<String> y) Train the model using labeled dataset
predict(Float64List x) Predict the label for a new input sample
decisionFunction(Float64List x) Return class scores (margin values)
classes Returns list of learned class labels
weights Access learned weights
biases Access learned intercepts

🧰 Utility Functions

  • normalize(List<Float64List> X) β†’ Min-max scaling to 0, 1
  • shuffleInPlace(List data, List labels, {int? seed}) β†’ Random shuffle with labels synced

πŸ”¬ Performance Notes

  • Designed for small to medium-sized datasets
  • Optimized for embedded & offline ML
  • Uses CPU-friendly gradient descent (not SMO or QP)
  • Not suitable for kernel methods or large-scale vector spaces

πŸ“Œ Use Cases

  • πŸŽ“ Teaching & academic learning in Dart
  • πŸ“± Lightweight ML inside Flutter apps
  • πŸ” Edge/IoT classification without TensorFlow
  • βš™οΈ Dart-native AI pipelines and decision systems

πŸ§ͺ Testing

Run all tests:

dart test

Test coverage includes:

  • Model training and prediction
  • Class scores with decisionFunction
  • Label consistency & structure

πŸ”— Author & Contributions

Created with ❀️ by Mehmet Γ‡elik
If you found this useful, consider ⭐ starring the project and following for more!

Contributions are welcome! Open issues, suggest features, or submit PRs.

πŸ“„ License

This project is licensed under the MIT License.
See the LICENSE file for details.

Libraries

ml_svm