Design and Evaluation of Analytical Models for Accelerating Decision-Making Processes in Big Data Analysis Using Machine Learning Techniques

Abstract

Machine learning models, particularly deep learning, play a crucial role in analyzing complex and large-scale data. This paper introduces a hybrid analytical framework combining two techniques, Annealed Gradient Descent (AGD) and Hybrid Orthogonal Projection and Estimation (HOPE), to improve prediction accuracy, reduce training time, and enhance stability in deep learning models. AGD, as an optimization algorithm, aims to accelerate convergence and prevent local minima trapping, thus increasing the training speed. Meanwhile, HOPE, through orthogonal projection, helps in dimensionality reduction and noise elimination. The proposed model, by leveraging these two techniques, significantly improves the performance of deep learning models in analyzing complex data and time series. Experimental results show that the proposed model can reduce training time by up to 40% and increase prediction accuracy by up to 6% compared to traditional methods such as SGD, SVM, and XGBoost.Furthermore, the hybrid AGD–HOPE framework demonstrates improved robustness and stability across diverse datasets and network architectures. The results highlight its effectiveness in handling high-dimensional and noisy data while maintaining consistent convergence behavior. These advantages make the proposed approach a promising solution for large-scale and time-series deep learning applications.