基于改进极限学习机的疾病预测研究

Research on disease prediction based on improved extreme learning machine

为了提高疾病预测的准确性,建立准确的疾病辅助诊断系统,为疾病诊断提供高可靠性指导意见,建立了改进极限学习机的疾病诊断模型。传统的极限学习机(extreme learning machine, ELM)随机获取初始权值和阈值,模型的预测精度有待提高。基于蚁群算法(ant colony optimization,ACO)强大的全局寻优能力,将极限学习机的初始权值和阈值作为蚁群算法的每只蚂蚁,进行最优路径规划,获得全局最优解即为优化后的极限学习机的权值和阈值。将所提方法与ELM,ACO,PSO算法在UCI数据上分别进行了仿真对比,实验结果显示无论在预测精度还是收敛速度上,所提方法均高于其他传统方法。在Pima Indian Diabetes和Breast Cancer的疾病诊断种,所提方法的预测精度也明显高于其他传统预测方法。验证了所提方法具有较高的疾病预测精度,适用于常见疾病的自动诊断。

In order to improve the accuracy of disease prediction, establish an accurate auxiliary diagnosis system, and provide high reliability guidance for disease diagnosis, a disease diagnosis model with improved extreme learning machine is established. Due to the random acquisition of initial weights and thresholds, the prediction accuracy of the traditional extreme learning machine(ELM) model needs to be improved. Since the ant colony optimization(ACO) has a strong global optimization ability, the initial weight and threshold of the ultimate learning machine are taken as each ant of the Ant colony optimization algorithm to conduct optimal path planning, and the global optimal solution is the weight and threshold of the optimized ultimate learning machine. The proposed method is compared with ELM, ACO and PSO algorithms on UCI data, and the experimental results show that the proposed method is higher than other traditional methods in both prediction accuracy and convergence speed. In the diagnosis of Pima Indian Diabetes and Breast Cancer, the prediction accuracy of the proposed method is also significantly higher than that of other traditional methods. It is verified that the method proposed has high accuracy of disease prediction and is suitable for automatic diagnosis of common diseases.