Multistrategy Improved Sparrow Search Algorithm Optimized Deep Neural Network for Esophageal Cancer

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Biology

Background:

• Deep neural networks (DNNs) excel at pattern recognition but struggle with small datasets.
• Shallow neural networks often outperform DNNs on limited data.
• Optimizing DNNs requires efficient search algorithms for connection weights.

Purpose of the Study:

• To develop an optimized deep neural network (DNN) model for predicting esophageal cancer survival rates.
• To enhance the sparrow search algorithm (SSA) for improved performance in DNN optimization.
• To evaluate the efficacy of the proposed model against existing algorithms.

Main Methods:

• A novel sparrow search algorithm (SSA) was developed, incorporating iterative map, iterative perturbation, and Gaussian mutation.
• The enhanced SSA was employed to optimize the connection weights of a deep neural network.
• The proposed model was implemented and tested on an esophageal cancer dataset, compared against six other algorithms.

Main Results:

• The enhanced sparrow search algorithm demonstrated superior search accuracy and convergence speed on fourteen benchmark functions.
• The optimized DNN model achieved a score of 0.92 across all eight scoring criteria on the esophageal cancer dataset.
• The proposed model significantly outperformed the other six comparative algorithms.

Conclusions:

• The improved sparrow search algorithm effectively optimizes deep neural networks for complex pattern recognition tasks.
• The developed deep neural network model offers a promising and effective approach for predicting esophageal cancer survival rates.
• This research highlights the potential of enhanced metaheuristic algorithms in medical data analysis and prediction.