Abstract:

Conventional support vector machines （SVMs） have the demerits ot low training speect and htgn computauonal requirements. To overcome the shortcomings, this paper applies the least squares support vector machine （LSSVM） to the issue of pattern classification. Meanwhile, this paper briefly points ouut the limitations of the decision directed acyclic graph （DDAG） algorithm, and uses the adaptive directed acyclic graph （ADAG） algorithm for solvinng multiclass problems. In order to enhance the learning speed of samples, this paper introduces sequential minimal optimization （SMO） to LSSVM, and f finally constructs the ADAGLSSVM algorithm. Owing to the lacking sparsity in the LSSVM algorithm, support vectors are pruned in the experiment to improve the accuracy and speed of classifiers. Experimental result shows that the performance of classifiers is improved after pruning.

Key words: least squares support vector machine, sequential minimal optimization, adaptive directed aeyclic graph