A New Neural Dynamic Classification Algorithm.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Pattern Recognition

Background:

• Effective classification algorithms require optimal feature spaces with large inter-cluster margins and minimal feature sets.
• Existing methods like Probabilistic Neural Networks (PNN) and Support Vector Machines (SVM) have limitations in feature space optimization.

Purpose of the Study:

• To introduce a novel supervised classification algorithm, Neural Dynamic Classification (NDC).
• To optimize feature space discovery and determine the minimum effective feature set for accurate classification.
• To leverage the Adeli and Park robust neural dynamic optimization model.

Main Methods:

• Development of the Neural Dynamic Classification (NDC) algorithm.
• Comparative analysis against Probabilistic Neural Network (PNN), Enhanced PNN (EPNN), and Support Vector Machine (SVM).
• Evaluation on standard benchmark datasets and the large-scale Mixed National Institute of Standards and Technology (NIST) handwritten digits database.

Main Results:

• NDC demonstrated superior classification accuracy compared to PNN, EPNN, and SVM across benchmark problems.
• EPNN also showed strong performance, generally ranking second to NDC.
• NDC exhibited smooth convergence curves, indicating algorithmic robustness and reliable performance.

Conclusions:

• Neural Dynamic Classification (NDC) is a highly effective algorithm for maximizing prediction accuracy.
• The algorithm successfully identifies optimal feature spaces and the necessary number of features for precise classification.
• NDC offers a robust and high-performing solution for complex classification tasks.