Elhoseny Mohamed, Metawa Noura, El-Hasnony Ibrahim M
Mansoura University, Egypt.
College of Computing and Informatics, University of Sharjah, the United Arab Emirates.
Sustain Comput. 2022 Sep;35:100778. doi: 10.1016/j.suscom.2022.100778. Epub 2022 Jun 17.
Global crises such as the COVID-19 pandemic and other recent environmental, financial, and economic disasters have weakened economies around the world and marginalized efforts to build a sustainable economy and society. Financial crisis prediction (FCP) has a significant impact on the economy. The growth and strength of a country's economy can be gauged by accurately predicting how many companies will fail and how many will succeed. Traditionally, there have been a number of approaches to achieving a successful FCP. Despite this, there is a problem with the accuracy of classification and prediction and with the legality of the data that is being used. Earlier studies have focused on statistical, machine learning (ML), and deep learning (DL) models to predict the financial status of a company. One of the biggest limitations of most machine learning models is model training with hyper-parameter fine-tuning. With this motivation, this paper presents an outlier detection model for FCP using a political optimizer-based deep neural network (OD-PODNN). The OD-PODNN aims to determine the financial status of a firm or company by involving several processes, namely preprocessing, outlier detection, classification, and hyperparameter optimization. The OD-PODNN makes use of the isolation forest (iForest) based outlier detection approach. Moreover, the PODNN-based classification model is derived, and the DNN hyperparameters are fine-tuned to boost the overall classification accuracy. To evaluate the OD-PODNN model, three different datasets are used, and the outcomes are inspected under varying performance measures. The results confirmed the superiority of the proposed OD-PODNN methodology over recent approaches.
新冠疫情等全球危机以及近期其他环境、金融和经济灾难削弱了全球经济，并使建设可持续经济和社会的努力边缘化。金融危机预测（FCP）对经济有着重大影响。通过准确预测有多少公司会失败以及有多少公司会成功，可以衡量一个国家经济的增长和实力。传统上，有多种方法可实现成功的金融危机预测。尽管如此，在分类和预测的准确性以及所使用数据的合法性方面仍存在问题。早期研究主要集中在统计、机器学习（ML）和深度学习（DL）模型上，以预测公司的财务状况。大多数机器学习模型最大的局限性之一是通过超参数微调进行模型训练。出于这一动机，本文提出了一种基于政治优化器的深度神经网络（OD-PODNN）用于金融危机预测的异常值检测模型。OD-PODNN旨在通过多个过程来确定公司的财务状况，这些过程包括预处理、异常值检测、分类和超参数优化。OD-PODNN利用基于孤立森林（iForest）的异常值检测方法。此外，推导了基于PODNN的分类模型，并对深度神经网络的超参数进行微调以提高整体分类准确率。为了评估OD-PODNN模型，使用了三个不同的数据集，并在不同的性能指标下检查结果。结果证实了所提出的OD-PODNN方法相对于近期方法的优越性。
