Quite a good number of population-based meta-heuristics based on mimicking natural phenomena are observed in the literature in resolving varieties of complex optimization problems. They are widely used in search of the optimal model parameters of artificial neural networks (ANNs). However, efficiencies of these are mostly dependent on fine tuning algorithm-specific parameters. Rao algorithms are metaphor-less meta-heuristics which do not need any algorithm-specific parameters. Functional link artificial neural network (FLANN) is a flat network and possesses the ability of mapping input–output nonlinear relationships by using amplification in input vector dimension. This article attempts to observe the efficacy of Rao algorithms on searching the most favorable parameters of FLANN, thus forming hybrid models termed as Rao algorithm-based FLANNs (RAFLANNs). The models are evaluated on forecasting five stock markets such as NASDAQ, BSE, DJIA, HSI, and NIKKEI. The RAFLANNs performances are compared with that of variations of FLANN (i.e., FLANN based on gradient descent, multi-verse optimizer, monarch butterfly optimization and genetic algorithm) and conventional models (i.e., MLP, SVM and ARIMA). The proposed models are found better in terms of prediction accuracy, computation time and statistical significance test.

在解决各种复杂优化问题的文献中，观察到大量基于模仿自然现象的基于群体的元启发式算法。它们被广泛用于寻找人工神经网络 (ANN) 的最佳模型参数。然而，这些的效率主要取决于微调特定于算法的参数。Rao 算法是无隐喻的元启发式算法，不需要任何特定于算法的参数。功能链接人工神经网络（FLANN）是一种扁平网络，具有通过使用输入向量维度的放大来映射输入-输出非线性关系的能力。本文试图观察 Rao 算法在搜索 FLANN 最有利参数方面的效果，因此形成称为基于 Rao 算法的 FLANNs (RAFLANNs) 的混合模型。这些模型是在预测 NASDAQ、BSE、DJIA、HSI 和 NIKKEI 等五个股票市场上进行评估的。RAFLANN 的性能与 FLANN（即基于梯度下降、多节优化器、帝王蝶优化和遗传算法的 FLANN）和传统模型（即 MLP、SVM 和 ARIMA）的变体的性能进行了比较。发现所提出的模型在预测精度、计算时间和统计显着性检验方面更好。帝王蝶优化和遗传算法）和传统模型（即 MLP、SVM 和 ARIMA）。发现所提出的模型在预测精度、计算时间和统计显着性检验方面更好。帝王蝶优化和遗传算法）和传统模型（即 MLP、SVM 和 ARIMA）。发现所提出的模型在预测精度、计算时间和统计显着性检验方面更好。