Robust Least-Squares Support Vector Machine Using Probabilistic Inference.

Area of Science:

• Machine Learning
• Statistical Modeling

Background:

• Least-squares support vector machine (LS-SVM) is widely used but struggles with non-Gaussian noise.
• Existing LS-SVM methods lack robustness when dealing with data contaminated by non-Gaussian noise.

Purpose of the Study:

• To propose a novel probabilistic LS-SVM to enhance modeling reliability for data affected by non-Gaussian noise.
• To develop a robust machine learning model capable of handling noise in datasets.

Main Methods:

• Analyzed and estimated the stochastic effect of noise on kernel functions and regularization parameters.
• Constructed a new objective function in a probabilistic sense.
• Developed a probabilistic inference method to estimate the distribution of model parameters, including kernel and regularization parameters.
• Created a solving strategy utilizing distribution information for the new objective function.

Main Results:

• The proposed probabilistic LS-SVM demonstrates enhanced modeling reliability compared to the deterministic LS-SVM.
• The method effectively models noise data by establishing a distribution relationship between the model and noise.
• Validated effectiveness using both artificial and real-world datasets.

Conclusions:

• The probabilistic LS-SVM offers a more robust approach for modeling data contaminated with non-Gaussian noise.
• This probabilistic framework improves the reliability and performance of LS-SVM in challenging data scenarios.