Neural Network Research Articles

Physics embedded neural network: Novel data-free approach towards scientific computing and applications in transfer learning

This paper introduces a novel approach for solving partial differential equations using neural networks for scientific computing applications. The proposed approach is referred to as physics embedded neural network (or PENN) due to its unique modified architecture designed to contain the partial differential equation information within the neural network, which is built upon a feed-forward neural network (NN) with fully connected layers. The key aspect of PENN is the numerical embedding of a differential equation associated with physical problem in the penultimate layer of the neural network. Here we propose that through such integration of PDE to the neural network’s penultimate layer, we can produce a fast numerical solution of the PDE, and with time performance comparable with Finite Element Method (FEM). Due to direct solution of the PDE, the NN solution occurs without training and testing phases which are normally applied when using NN in a data-based training. To demonstrate the effectiveness of the proposed architecture, we investigate variety of cases focused around second order in-homogeneous PDE in 2 dimensional (2D) problems. Additionally, we demonstrate the effectiveness of our approach by applying it to solve electromagnetic problem to illustrate its engineering applications. In this study, we also conducted a comparison of solution time and error achieved using our proposed network with a classical finite element method (FEM) solver. The convergence was found exponential with number of epochs proving an error significantly lower error compared to FEM and reported in other NN based algorithms such as Physics Informed Neural Networks. Furthermore, we explored the application of PENN in transfer learning. Our findings showed that by using transfer learning, up to 5 times additional reduction in simulation time is possible as compared to a direct solution through PENN. Overall, the study contributes to the development of efficient and accurate methods for solving PDEs through neural network optimization and provides important insights into the potential benefits of incorporating transfer learning techniques in numerical modeling of physical phenomenology.

A multivariable adaptive control method for aeroengine with H∞ performance considering engine output limitation protection based on fully adjustable Neural Network

Aeroengines, being highly nonlinear systems subject to external disturbances, present a challenge for conventional control methods in achieving satisfactory performance. The radial basis function neural networks possess the capability to effectively model highly approximate nonlinear functions, making them suitable for implementation in aeroengine control. However, the radial basis function (RBF) neural network is a local approximation model that inadequately incorporates historical information. Additionally, the tracking error of neural networks can also impact control effectiveness. To address this issue, this paper proposes a multivariable control method for aeroengines with H∞ performance based on fully adjustable Long Short-term Memory-Radial Basis Function Neural Network (LSTM-RBFNN). Firstly, a factual mathematical model of an aeroengine is established based on the principles of aerodynamics and thermodynamics. Secondly, the paper proposed an LSTM-RBF neural network architecture to enhance prediction performance and proposed using a finite time perturbation estimator to gauge the approximate error of neural networks, thereby acquiring immeasurable approximative error information. Then, in order to mitigate the influence of neural network tracking error on the loop, the estimated errors are incorporated into the primary loop controller, and a static state disturbance full feedback H∞ controller is proposed. Then, leveraging the LSTM-RBF architecture proposed in this study, we further employ Taylor expansion of nodes to derive the adaptive law of the neural network and propose a novel neural network adaptive law that incorporates tracking error estimated by the disturbance observer into the adjustment mechanism to enhance tracking performance. The stability of this adaptive law has been proven based on Lyapunov function analysis. And considering the output limitation protection problem of aeroengine, a virtual command value based limitation protection method is proposed for the above control method, and it is proved that this limitation method does not damage the stability and design process of the original control method. Finally, numerical simulation and hardware in loop simulation experiments were conducted, and the results showed that under the same adaptive gain of the neural network, the performance indicators (IAE, IATE, σ) could be reduced by more than 50% compared to other methods, which demonstrates superior tracking and disturbance rejection performance achieved by our proposed method.

Assessment of matting agent effect on polyurethane coatings' perceived blackness through geometric properties modeling: An AI approach

This study investigates the effect of matting agent on the perceived blackness of polyurethane automotive finishes by examining their impact on geometric properties using an AI-based neural network model. Black polyurethane coatings were prepared with varying gloss levels by incorporating different amounts of matting agent. Observers ranked these samples based on perceived blackness using the pair comparison method. The data features included the colorimetric properties and geometric attributes of the black polyurethane panels, with visual scales obtained as target values. Various neural networks were employed: Single-Layer Neural Network for linear regression, Multi-Layer Neural Network, Deep Neural Network, and Dropout-Enhanced Deep Neural Network for non-linear regression. Feature selection was applied to eliminate irrelevant or redundant features. The findings indicated that the Multi-Layer Neural Network for non-linear regression, utilizing geometric features, achieved favorable evaluation metrics, including mean squared errors, and R2 values. Additionally, this model predicted the visual scale values more accurately compared to other neural network models. By simulating this behavior, the research effectively eliminates the need for time-consuming visual assessment experiments traditionally used for blackness evaluations in industrial settings. This method greatly simplifies the evaluation process, conserving both time and resources while maintaining reliable and precise assessments.

Neural networks through the lens of evolutionary dynamics.

This article revisits Artificial Neural Networks (NNs) through the lens of Evolutionary Dynamics. The two most important features of NNs are shown to reflect the two most general processes of Evolutionary Dynamics. This overlap may serve as a new and powerful connection between NNs and Evolutionary Dynamics, which encompasses a body of knowledge that has been built over multiple centuries and has been expanded to inspire applications across a vast range of disciplines. Consequently, NNs should also be applicable across the same range of disciplines-that is, much more broadly than initially envisioned. The article concludes by opening questions about NN dynamics, based on the new connection to Evolutionary Dynamics.

A comparison of reinforcement learning policies for dynamic vehicle routing problems with stochastic customer requests

This paper presents directions for using reinforcement learning with neural networks for dynamic vehicle routing problems (DVRPs). DVRPs involve sequential decision-making under uncertainty where the expected future consequences are ideally included in current decision-making. A frequently used framework for these problems is approximate dynamic programming (ADP) or reinforcement learning (RL), often in conjunction with a parametric value function approximation (VFA). A straightforward way to use VFA in DVRP is linear regression (LVFA), but more complex, non-linear predictors, e.g., neural network VFAs (NNVFA), are also widely used. Alternatively, we may represent the policy directly, using a linear policy function approximation (LPFA) or neural network PFA (NNPFA). The abundance of policies and design choices complicate the use of neural networks for DVRPs in research and practice. We provide a structured overview of the similarities and differences between the policy classes. Furthermore, we present an empirical comparison of LVFA, LPFA, NNVFA, and NNPFA policies. The comparison is conducted on several problem variants of the DVRP with stochastic customer requests. To validate our findings, we study realistic extensions of the stylized problem on (i) a same-day parcel pickup and delivery case in the city of Amsterdam, the Netherlands, and (ii) the routing of robots in an automated storage and retrieval system (AS/RS). Based on our empirical evaluation, we provide insights into the advantages and disadvantages of neural network policies compared to linear policies, and value-based approaches compared to policy-based approaches.

Fast gradient-free activation maximization for neurons in spiking neural networks

Elements of neural networks, both biological and artificial, can be described by their selectivity for specific cognitive features. Understanding these features is important for understanding the inner workings of neural networks. For a living system, such as a neuron, whose response to a stimulus is unknown and not differentiable, the only way to reveal these features is through a feedback loop that exposes it to a large set of different stimuli. The properties of these stimuli should be varied iteratively in order to maximize the neuronal response. To utilize this feedback loop for a biological neural network, it is important to run it quickly and efficiently in order to reach the stimuli that maximizes certain neurons’ activation with the least number of iterations possible. Here we present a framework with an efficient design for such a loop. We successfully tested it on an artificial spiking neural network (SNN), which is a model that simulates the asynchronous spiking activity of neurons in living brains. Our optimization method for activation maximization is based on the low-rank Tensor Train decomposition of the discrete activation function. The optimization space is the latent parameter space of images generated by SN-GAN or VQ-VAE generative models. To our knowledge, this is the first time that effective AM has been applied to SNNs. We track changes in the optimal stimuli for artificial neurons during training and show that highly selective neurons can form already in the early epochs of training and in the early layers of a convolutional spiking network. This formation of refined optimal stimuli is associated with an increase in classification accuracy. Some neurons, especially in the deeper layers, may gradually change the concepts they are selective for during learning, potentially explaining their importance for model performance. The source code of our framework, MANGO (for Maximization of neuronal Activation via Non-Gradient Optimization) is available on GitHub (https://github.com/iabs-neuro/mango)

The Secure Therapeutic Effects of Recently Developed Antipsychotic Drugs and Updated Neural Networks in Schizophrenia

Introduction: Schizophrenia and schizoaffective disorder are treated in most cases with antipsychotic drugs of the second generation. These drugs block dopaminergic and serotonergic receptors, i.e., D2 and 5-HT2A receptors, and cause different adverse effects, for example, movement disturbances of the extrapyramidal system and adverse effects of vital parameters and of the heart. These drugs treat positive symptoms in schizophrenia and, to a lesser extent, negative symptoms. This review presents the development of newer antipsychotic drugs. Methods/Material: References were taken from PubMed after using the following keywords: schizophrenia, schizoaffective disorder, antipsychotic drug, neurotransmitter and neuropeptide. Among these newer antipsychotic drugs are cariprazine, brexipiprazole and lumateperone, which exert a partial agonistic effect at D2 and 5-HT2A receptors, pimavanserin, a 5-HT2A receptor antagonist which treats negative symptoms in schizophrenia as an add-on therapy, olanzapine combined with samidorphan, which reduces weight gain, and M4 or M1 receptor agonists, for example, xanomeline with an antipsychotic effect combined with trospium, an anticholinergic drug. Neural networks were updated in order to deduce the antipsychotic mechanism of action of newer antipsychotic drugs, especially xanomeline. Results: The newer antipsychotic drugs cariprazine, brexipiprazole and lumateperone show antipsychotic, antimanic and anti-depressive effects, however, the efficacy on psychotic symptoms in long-term treatment has not yet been examined. Pimavanserin reduces negative symptoms in schizophrenia as an additional pharmacotherapy to treat this disorder. Olanzapine combined with samidorphan exerts good antipsychotic effects and reduces weight gain. The new antipsychotic drug xanomeline, the antipsychotic effect of which is quite different from the antidopaminergic effect, treats positive and negative symptoms in schizophrenia. Its mechanism of action was deduced from the neural networks presented. The long-term efficacy should still be examined. Conclusion: This review is focused on newer antipsychotic drugs. The long-term efficacy of cariprazine, brexipiprazole and lumateperone in the treatment of schizophrenia should be examined furthermore. Neural networks in the brain areas involved in schizophrenia should be examined and updated furthermore. Newer antipsychotic drugs, for example, xanomeline, an M4 or M1 receptor agonist, which has been combined with trospium, an anticholinergic drug, the mechanism of action of which can be derived from the neural network suggested in this review.

Predicting mortality in upper gastrointestinal tract cancer using long short-term memory neural networks.

340 Background: Upper gastrointestinal (GI) tract cancers represent a significant global health burden, characterized by high morbidity and mortality rates, with an increasing incidence among aging populations and varying outcomes based on gender and ethnicity. Early prediction of mortality rates in upper gastrointestinal cancer can significantly enhance treatment strategies and patient outcomes. This study describes current trends in upper gastrointestinal cancer and proposes a predictive model, developed using Long Short-Term Memory (LSTM) neural networks to project future trends using demographic data. Methods: We extracted mortality data from the Center for Disease Control and Prevention (CDC) Wide-Ranging Online Data for Epidemiological Research (WONDER) Database spanning 1999, normalized it and encoded categorical variables such as age, sex, ethnicity, and race numerically. Using Studio R to analyze mortality trends, we developed an LSTM model known for its ability to capture and utilize long term data patterns. The model, trained with a sequence length of 20, predicted future data points based on the previous 20. We then assessed its accuracy and generalization with a test set. Ultimately, our model is able to utilize demographic variables to forecast annual mortality and survival rates. Results: The mortality rates for upper GI tract cancers are notably higher among elderly individuals, particularly those aged 80 and above, with males and certain ethnic groups, such as Hispanics. The accuracy of the trained model was evaluated by calculating the Mean Squared Error (MSE), a measure of the difference between the predicted and actual value, with zero indicating no difference between the two. Our LSTM model achieved a training MSE of 0.00085 and a validation MSE of 0.0016, indicating good reliability and accuracy. Conclusions: Our analysis highlighted clear mortality trends and demonstrated the model’s effectiveness in improving mortality predictions in upper gastrointestinal cancer. However, limitations due to the dataset’s lack of clinical variables (such as comorbidities, cancer stage, and prior treatments) and its reliance on a single dataset may affect external validity. Future research is needed to determine if the model can be applied for clinical prognosis or strategic planning. We ultimately created and validated an LSTM neural network model with the CDC WONDER dataset, showing promising results. Future development will aim to include clinical characteristics to create a more personalized predictive tool.

Prediction of ketosis using radial basis function neural network in dairy cattle farming.

The purpose of the paper was to apply an Artificial Neural Networks with Radial Basis Function to develop an application model for diagnosing a subclinical ketosis type I and II in dairy cattle. While building the neural network model, applied methodology was compatible to the procedures used in Data Mining processes.The data set was created based on the composition of milk samples of 1520 Polish Holstein-Friesian cows. The milk samples were collected during test-day milkings and made available by Polish Federation of Cattle Breeders and Milk Producers. The milk composition parameters were used as the input variables for RBF network models. The value of the output variable was determined based on the content of β-hydroxybutyric acid in blood of cows. In the next stage of the work, the qualities of the pre-selected models were compared and the best ones were chosen. The sensitivity and specificity as well as the size of the AUC (Area Under the Curve) under the ROC (Receiver Operating Characteristic) were taken as the main criteria for network models evaluation. The model characterized by sensitivity of 0.86, specificity of 0.71 and AUC of 0.89 was selected for ketosis type I. The optimal for ketosis type II showed the sensitivity and specificity 0.81 and 0.75, respectively, and the size of AUC above 0.85. Chosen models were recorded using the predictive modelling markup language (PMML) for data mining models to be shared and used between the different applications.

Hessian-based mixed-precision quantization with transition aware training for neural networks.

Model quantization is widely used to realize the promise of ubiquitous embedded deep network inference. While mixed-precision quantization has shown promising performance, existing approaches often rely on time-consuming search process to determine the optimal bit configuration. To address this, we introduce Hessian-based Mixed-Precision Quantization Aware Training(HMQAT) to decrease the search overhead of bit configuration. By using the sensitivity metric of joint average Hessian trace and parameter size, HMQAT effectively guides the search process. We solve the optimization problem about bit configuration automatically using a Pareto frontier method. The lowest search overhead can be achieved through our scheme. Additionally, our approach incorporates quantization transition aware fine-tuning of scale factor. This strategy consistently ensures optimal inference performance along the accuracy-size Pareto frontier across multiple models. We extensively evaluated our method on ImageNet and CIFAR10. In particular, we show that compared to the baseline, HMQAT achieves a 10.34× reduction in model size while retaining 99.81% of the Top-1 accuracy on ResNet18 for ImageNet. Moreover, HMQAT surpassed the state-of-the-art mixed-precision quantization methods in compressing neural networks with reduced search cost and achieving a satisfying trade-off between size and accuracy. This study paves the way of deploying neural networks on lightweight devices.

A neural network integrated mathematical model to analyse the impact of nutritional status on cognitive development of child

Cognitive development is a crucial developmental aspect of children. It is a concise field of study in psychology and neuroscience that focuses on various developmental aspects of the brain. Among all other factors, nutritional status is believed to play a very important role in cognitive development. The purpose of this work is to analyze the impacts of different nutritional status levels on the child’s cognitive development. This work designs a model that uses a neural network and differential equations. The neural network is applied on a dataset called “Child Birth Weight Dataset” available at IEEE Dataport ( http://dx.doi.org/10.21227/dvd4-3232) for finding the nutritional status of a child. The different levels of nutritional status, such as low-nutritional status, normal-nutritional status, and over-nutritional status are integrated with the formulated differential equations. The model is computationally simulated considering four different sets of parameter values that represent four different perspectives such as ‘only positive’, ‘only negative’, ‘mix and unequal weight’, and ‘mix and equal weight’ of the influencing factors. The experimental results show that normal-nutritional status is the best nutritional status for cognitive development. However, the best cognitive development happens when all other influencing factors like environmental effects, socioeconomic status, heredity, learning opportunities, and use of experiences are given equal importance. The results also depict that the low- and over-nutritional status cannot restrict cognitive development for a long time. After a certain period, the development gets triggered and it happens. It may be slow and not up to the mark of the development under normal-nutritional status, but it happens. Simply it can be said that nutritional status alone does not have control over the cognitive development of a child. Along with nutritional status, other influencing factors are important too.

Probabilistic prediction of phosphate ion adsorption onto biochar materials using a large dataset and online deployment.

Phosphate (PO4(III)) contamination in water bodies poses significant environmental challenges, necessitating efficient and accurate methods to predict and optimize its removal. The current study addresses this issue by predicting the adsorption capacity of PO4(III) ions onto biochar-based materials using five probabilistic machine learning models: eXtreme Gradient Boosting LSS (XGBoostLSS), Natural Gradient Boosting, Bayesian Neural Networks (NN), Probabilistic NN, and Monte-Carlo Dropout NN. Utilizing a dataset of 2952 data points with 16 inputs, XGBoostLSS demonstrated the highest R2 (0.95) on new adsorbents. SHapely Additive exPlanations analysis showed that adsorption experimental conditions had the most significant impact (43%), followed by synthesis conditions (29%) and adsorbent characteristics (28%). Optimized conditions included an initial PO4(III) concentration of 125mg/L, carbon content of 11.5%, oxygen content of 23%, a contact time of 1440min, a heating rate of 5°C/min, 200rpm, and a surface area of 410m2/g, using Ra-LDO adsorbent synthesized from rape cabbage feedstock. This study developed and presented a practical online framework for predicting PO4(III) removal onto biochar using a web-based graphical user interface.

Learning a more compact representation for low-rank tensor completion

Transform-based tensor nuclear norm (TNN) methods have gained considerable attention for their effectiveness in addressing tensor recovery challenges. The integration of deep neural networks as nonlinear transforms has been shown to significantly enhance their performance. Minimizing transform-based TNN is equivalent to minimizing the ℓ1 norm of singular values in the transformed domain, which can be interpreted as finding a sparse representation with respect to the bases supported by singular vectors. This work aims to advance deep transform-based TNN methods by identifying a more compact representation through learnable bases, ultimately improving recovery accuracy. We specifically employ convolutional kernels as these learnable bases, demonstrating their capability to generate more compact representation, i.e., sparser coefficients of real-world tensor data compared to singular vectors. Our proposed model consists of two key components: a transform component, implemented through fully connected neural networks (FCNs), and a convolutional component that replaces traditional singular matrices. Then, this model is optimized using the ADAM algorithm directly on the incomplete tensor in a zero-shot manner, meaning all learnable parameters within the FCNs and convolution kernels are inferred solely from the observed data. Experimental results indicate that our method, with this straightforward yet effective modification, outperforms state-of-the-art approaches on video and multispectral image recovery tasks.

Automated identification of impact spatters and fly spots with a residual neural network.

In criminal investigations, distinguishing between impact spatters and fly spots presents a challenge due to their morphological similarities. Traditional methods of bloodstain pattern analysis (BPA) rely significantly on the expertise of professional examiners, which can result in limitations including low identification efficiency, high misjudgment rates, and susceptibility to external disturbances. To enhance the accuracy and scientific rigor of identifying impact spatters and fly spots, this study employed artificial intelligence techniques in image recognition and transfer learning. Two types of bloodstains obtained from simulation experiments were utilized as datasets, and a pre-trained neural network, ResNet-18, was employed for feature extraction. The original fully connected layer was replaced, and a new fully connected layer with a dimensionality of 2 was introduced to fulfil the task requirements. The results demonstrate that the transfer learning network model, based on ResNet-18, achieved a maximum accuracy of 93 % in morphologically identifying impact spatters and fly spots. The objective is to assist crime scene investigators and BPA analysts to identify bloodstains at homicide scenes conveniently, rapidly and accurately, thereby furnishing scientific evidence for scene reconstruction and advancing BPA toward intelligent practices.

Groundwater level forecasting using empirical mode decomposition and wavelet-based long short-term memory (LSTM) neural networks

Groundwater level forecasting using empirical mode decomposition and wavelet-based long short-term memory (LSTM) neural networks

Imaging anisotropic conductivity from internal measurements with mixed least-squares deep neural networks

In this work we develop a novel algorithm, termed as mixed least-squares deep neural network (MLS-DNN), to recover an anisotropic conductivity tensor from the internal measurements of the solutions. It is based on applying the least-squares formulation to the mixed form of the elliptic problem, and approximating the internal flux and conductivity tensor simultaneously using deep neural networks. We provide error bounds on the approximations obtained via both population and empirical losses. The analysis relies on the canonical source condition, approximation theory of deep neural networks and statistical learning theory. We also present multiple numerical experiments to illustrate the performance of the method, and conduct a comparative study with the standard Galerkin finite element method and physics informed neural network. The results indicate that the method can accurately recover the anisotropic conductivity in both two- and three-dimensional cases, up to 10% noise in the data.

Smart energy management: Process structure-based hybrid neural networks for optimal scheduling and economic predictive control in integrated systems

Integrated energy systems (IESs) are complex prosumers consisting of diverse operating units spanning multiple domains. The tight integration of these units results in varied dynamic characteristics and intricate nonlinear process interactions, making detailed dynamic modeling and successful operational optimization challenging. To address these concerns, we propose a process structure-based hybrid time-series neural network (NN) surrogate to predict the dynamic performance of IESs across multiple time scales. This neural network-based modeling approach develops time-series multi-layer perceptrons (MLPs) for the operating units and integrates them with prior process knowledge about system structure and fundamental dynamics. This integration forms three hybrid NNs—long-term, slow, and fast MLPs—that predict the entire system dynamics across multiple time scales. Leveraging these MLPs, we design an NN-based scheduler and an NN-based economic model predictive control (NEMPC) framework to meet global operational requirements: rapid electrical power responsiveness to operators’ requests, adequate cooling supply to customers, and increased system profitability, while addressing the dynamic time-scale multiplicity present in IESs. The proposed day-ahead scheduler is formulated using the ReLU network-based MLP, which effectively represents IES performance under a broad range of conditions from a long-term perspective. The scheduler is then exactly recast into a mixed-integer linear programming problem for efficient evaluation. The real-time NEMPC, based on slow and fast MLPs, comprises two sequential distributed control agents: a slow NEMPC for the cooling-dominant subsystem with slower transient responses and a fast NEMPC for the power-dominant subsystem with faster responses. These agents collaborate in the decision-making process to achieve dynamic synergy in real time while reducing computational costs. Extensive simulations demonstrate that the developed scheduler and NEMPC schemes outperform their respective benchmark scheduler and controller by about 25% and 40%. Together, they enhance overall system performance by over 70% compared to benchmark approaches.

A comprehensive dataset and neural network approach for named entity recognition in the Uzbek language.

In this study, the authors presented a dataset for named entity recognition in the Uzbek language. The dataset consists of 2000 sentences and 25,865 words, and the sources were legal documents and hand-crafted sentences annotated using the BIOES scheme. The study is complemented by the fact that the authors demonstrated the applications of the created dataset by training a language model using the CNN + LSTM architecture, which achieves high accuracy in NER tasks, with an F1 score of 90.8 %, precision of 93.9 %, and recall of 88.0 % on the test set. The proposed dataset and trained model contribute to the development of natural language processing in the Uzbek language. In addition, the authors also conducted an analysis of existing works, as well as a comparative analysis, which will help to identify the distinctive features and novelty of the proposed work. Moreover, in conclusion, the authors propose possible scenarios for the development of the work, in the form of further scaling of the dataset, as well as the use of other neural network architectures.

Study on prediction models of oxygenated components content in biomass pyrolysis oil based on neural networks and random forests

Study on prediction models of oxygenated components content in biomass pyrolysis oil based on neural networks and random forests

Can the number of confirmed COVID-19 cases be predicted more accurately by including lifestyle data? An exploratory study for data-driven prediction of COVID-19 cases in metropolitan cities using deep learning models.

The COVID-19 outbreak has significantly impacted human lifestyles and life patterns. Therefore, data related to human social life may tell us the increase or decrease in the number of confirmed COVID-19 cases. However, although the number of confirmed cases is affected by social life, it is difficult to find studies that attempt to predict the number of confirmed cases using various lifestyle data. This paper attempted an exploratory data analysis to see if the number of confirmed cases could be predicted more accurately by including various lifestyle data. We included taking public transportation, watching a movie at the cinema, and accommodation at a motel in the lifestyle data. Finally, a 'lifestyle addition' set was constructed that added lifestyle data to the number of past confirmed cases and search term frequency data. The deep learning algorithms used in the analysis are deep neural networks (DNNs) and recurrent neural networks (RNNs). Performance differences across data sets and between deep learning models were tested to be statistically significant. Among metropolitan cities in South Korea, Seoul (9.6 million) with the largest population and Busan (3.4 million) with the second largest population had the lowest error rate in 'lifestyle addition' set. When predicting with the 'lifestyle addition' set, in Seoul, the error rate was reduced to 20.1%, and in Busan, the graph of the actual number of confirmed cases and the predicted graph were almost identical. Through this study, we were able to identify three notable results that could contribute to predicting the number of patients infected with epidemic in the future.