Fast Support Vector Classification for Large-Scale Problems.

Area of Science:

• Machine Learning
• Computer Science
• Data Science

Background:

• Support Vector Machines (SVMs) are powerful for classification but struggle with large datasets due to high memory and time demands.
• Existing SVM implementations like Libsvm and Liblinear face performance limitations on extensive or complex datasets.

Purpose of the Study:

• To introduce the Fast Support Vector Classifier (FSVC), an optimized algorithm designed to overcome the scalability issues of traditional SVMs.
• To develop a machine learning model that is both computationally efficient and memory-sparing for large-scale data classification.

Main Methods:

• FSVC employs an efficient closed-form training process, eliminating iterative procedures.
• It utilizes a concise set of class prototypes, minimizing the need to store numerous support vectors.
• A novel method for direct radial basis function kernel spread selection from data, bypassing classifier execution and hyper-parameter tuning.

Main Results:

• FSVC demonstrates significantly lower memory requirements and faster processing times compared to Liblinear and Libsvm.
• The algorithm successfully processes datasets with millions of patterns, thousands of inputs, and over a hundred classes within hours.
• FSVC exhibits superior performance and scalability, even on large datasets where other methods fail due to memory constraints.

Conclusions:

• FSVC provides a highly efficient and scalable solution for large-scale machine learning classification tasks.
• The algorithm's predictable performance based on dataset size allows for accurate time estimation.
• FSVC is adaptable to systems with limited memory, enabling the classification of massive datasets.