Description:[500] Objective. Most infectious illnesses spread quickly. The goal of this research is to create a model that can use real hospital data along with information from several sources, such as weather and disease transmission patterns, to offer precise early warnings of infectious illnesses.

[501] Methods. A multi self-regression deep (MSRD) neural network was built and seven prevalent infectious illnesses in medical institutions were screened using daily data reported for infectious diseases that were gathered from many big general hospitals in China between 2012 and 2020. The model, which takes into consideration the historical development features in time-series data and takes into account the current influencing variables, may successfully simulate the epidemiological trend of infectious illnesses using a recurrent neural network as its fundamental structure. Mean absolute error (MAE) and root mean squared error were used to assess the model's fitting and prediction accuracy.

[502] Results. The suggested method greatly outperforms the susceptible-exposed-infected-removed (SEIR) infectious disease dynamics model since it tackles issues with hard-to-obtain quantitative data, such as latent population, overfitting of lengthy time series, and focusing solely on a single series of the number of ill individuals without taking into account the epidemiological features of infectious diseases. In this study, we also compare certain specific machine learning techniques. According to experimental data, the suggested strategy produces an MAE of 1.37882 for influenza and 0.69288 for hand, foot, and mouth disease, respectively.

[503] Conclusion. This research proposes an MRSD-based infectious disease prediction model that may offer precise forecasts for epidemic trends along with daily and immediate updates.

FIELD OF THE INVENTION
[504] Our Invention is related to a Computer Vision based Infection Control System in Hospital.

BACKGROUND OF THE INVENTION

[505] The direct reporting network encompassed 100% of disease control centers, 95% of medical and health institutions at the county level and above, and 70% of rural health facilities nationally as early as the end of 2006, which resulted in ten times faster reporting of infectious illnesses.

[506] COVID-19 revealed the system's shortcomings in 2020: it lacked intelligent analysis of predefined warnings and active alerting, which hinders timely diagnosis and accurate risk prevention and control and raises the possibility of underreporting or delayed reporting.

[507] The majority of the government's current alerts about contagious diseases are also cautions about the state of the country. As the situation at local medical institutions develops, it is challenging to give accurate feedback, and there is a dearth of effective early warning systems based on the information gathered from medical institutions.

[508] The deployment of hospitals for the purpose of preventing and controlling epidemics is thus not supported by the current system, which will significantly diminish the role of general hospitals as front-line institutions for these purposes.

[509] To forecast changes in infectious diseases in hospitals, we suggested using an MSRD model. The suggested method works better than the SEIR model, according to experimental data.

[510] Additionally, the SEIR model's limitations in predicting hospital infections were explained. Additionally, we examine a number of machine learning and neural network techniques, including SVM, Lasso regression, Bi-LSTM, DNN, and LSTM.

[511] The MSRD method performs better than the previously mentioned techniques. By using a temporal frame to extract training data features, MSRD avoids the overfitting issue brought on by lengthy time series and the real-world implementation of no flexibility.

[512] Furthermore, merging the result of every time step with the matching original input enhances the model's fitting performance. The model's prediction of the effect of prevalent infectious diseases is in line with the high incidence of infectious diseases that occur in reality.

[513] In order to facilitate quick reaction and decision-making, the model integrates data from external sources with hospital data through the use of crowdsourcing, medical record information, and climate. It also helps hospitals anticipate and detect infectious disease outbreaks.

[514] The model may be expanded to include infectious disease monitoring in various kinds of hospitals, which will assist to improve infectious illness prediction and surveillance, standardize infectious disease treatment, and provide dynamic early warning of infectious diseases.

OBJECTIVES OF THE INVENTION
1. The objective of the invention is to provide a multi self-regression deep (MSRD) neural network was built and seven prevalent infectious illnesses in medical institutions were screened using daily data reported for infectious diseases that were gathered from many big general hospitals in China between 2012 and 2020.

2. The objective of the invention is to provide a model, which takes into consideration the historical development features in time-series data and takes into account the current influencing variables, may successfully simulate the epidemiological trend of infectious illnesses using a recurrent neural network as its fundamental structure.

3. The objective of the invention is to provide a Mean absolute error (MAE) and root mean squared error were used to assess the model's fitting and prediction accuracy. The suggested method greatly outperforms the susceptible-exposed-infected-removed (SEIR) infectious disease dynamics model since it tackles issues with hard-to-obtain quantitative data.

4. The objective of the invention is to provide a latent population, overfitting of lengthy time series, and focusing solely on a single series of the number of ill individuals without taking into account the epidemiological features of infectious diseases.
5. The objective of the invention is to provide a we also compare certain specific machine learning techniques. According to experimental data, the suggested strategy produces an MAE of 1.37882 for influenza and 0.69288 for hand, foot, and mouth disease, respectively.

SUMMARY OF THE INVENTION
[515] There are two sources of the data. The first is the official monthly scientific statistics on public health from 2012 to 2017 provided by the Centers for Disease Control and Prevention (CDC).

[516] The second set of data comes from Peking University Third Hospital's daily inpatient and outpatient medical records between 2012 and 2020. The hospital data center provided the hospital data. The hospital data center includes an HBase column database, a distributed file system, and Hadoop architecture.

[517] The Hive data warehouse, which is capable of carrying out analytical calculations and data storage with ease. After in-depth mining of 110,000 historical data points on infectious diseases accumulated in a large hospital over eight years.

[518] The big data technology is applied to collect and clean the clinical data and then store and manage them in a centralized way, which provides the necessary basis for the training and application of an early-warning model for infectious diseases.

[519] Given the strong correlation between some viral illnesses and climatic conditions, this study also gathered daily temperature, humidity, wind, and other climate data from the national meteorological data department's website.

[520] The spread of infectious illnesses is strongly correlated with human activity in addition to the previously mentioned number of infectious diseases and environmental variables.

[521] For instance, infectious illnesses have a higher chance of spreading rapidly in settings where people congregate. As a result, this element is also utilized in this study's data feature for prediction, which is consistent with the research on a few infectious illnesses.

BRIEF DESCRIPTION OF THE DIAGRAM
Fig.1: Computer Vision based Infection Control System in Hospital Flow.
Fig.2: Computer Vision based Infection Control System in Hospital Process.
Fig.3: Computer Vision based Infection Control System in Hospital.
DESCRIPTION OF THE INVENTION

[522] We analyze seven common infectious diseases, including hand-foot-and-mouth disease (HFMD), influenza, viral hepatitis, infectious diarrheal disease, scarlet fever, syphilis, and tuberculosis during in this study.

[523] Our analysis is based on actual hospital outpatient data collected from the Peking University Third Hospital between 2012 and 2020. Of these, the prevalence of viral hepatitis and influenza is growing slowly each year, while that of infectious diarrheal disease, scarlet fever, syphilis, and tuberculosis has declined annually. This indicates that overall infectious disease prevention and control is effective.

[524] The government ought to take proactive precautions beforehand. More people in 2020 are impacted by the COVID-19 pandemic, and fewer people are really visiting hospitals.

[525] We built an MSRD model based on data analysis from 2012 to 2020, testing various combinations of parameters such as sliding window length candidates (3, 5, 7, 9, and 14), number of candidates for LSTM neurons (6, 8, 16, 32, 64, 128), number of candidates for feedforward neural network neurons (32, 64, 128, 256, 512), and multiple learning rates.

[526] The following are the ideal values for the aforementioned parameters: The length of the sliding window is 7.

[527] There are 16 LSTM neurons, 128 feedforward neural network neurons, and a 0.001 learning rate. The number of LSTM and feedforward neurons rose with the sliding window length, and this was shown to improve the MSRD model's performance in the training data.

[528] However, the test set's performance first climbed and then declined. The findings show that the model's ability to fit data is positively connected with both the model's complexity and the number of days of historical data employed.

[529] The generalization ability varies, into a trend of first increasing and then decreasing. The model developed to forecast the incidence of infectious diarrhea in 2021 is depicted in Figure 3 and was trained using the ideal parameters previously mentioned.

[530] According to the estimate, there will be fewer cases of infectious diarrhea in 2021. For the outcomes of the other infectious illness predictions, see Supplementary Material 1.
, Claims:I/WE CLAIMS
1) A multi self-regression deep (MSRD) neural network was built and seven prevalent infectious illnesses in medical institutions were screened using daily data reported for infectious diseases that were gathered from many big general hospitals in China between 2012 and 2020. The model, which takes into consideration the historical development features in time-series data and takes into account the current influencing variables, may successfully simulate the epidemiological trend of infectious illnesses using a recurrent neural network as its fundamental structure. Mean absolute error (MAE) and root mean squared error were used to assess the model's fitting and prediction accuracy. The suggested method greatly outperforms the susceptible-exposed-infected-removed (SEIR) infectious disease dynamics model since it tackles issues with hard-to-obtain quantitative data, such as latent population, overfitting of lengthy time series, and focusing solely on a single series of the number of ill individuals without taking into account the epidemiological features of infectious diseases. In this study, we also compare certain specific machine learning techniques. According to experimental data, the suggested strategy produces an MAE of 1.37882 for influenza and 0.69288 for hand, foot, and mouth disease, respectively.

2) According to claim1# the invention is to a multi self-regression deep (MSRD) neural network was built and seven prevalent infectious illnesses in medical institutions were screened using daily data reported for infectious diseases that were gathered from many big general hospitals in China between 2012 and 2020.

3) According to claim1,2# the invention is to a model, which takes into consideration the historical development features in time-series data and takes into account the current influencing variables, may successfully simulate the epidemiological trend of infectious illnesses using a recurrent neural network as its fundamental structure.

4) According to claim1,2,3,4# the invention is to a Mean absolute error (MAE) and root mean squared error were used to assess the model's fitting and prediction accuracy. The suggested method greatly outperforms the susceptible-exposed-infected-removed (SEIR) infectious disease dynamics model since it tackles issues with hard-to-obtain quantitative data.

5) According to claim1# the invention is to a latent population, overfitting of lengthy time series, and focusing solely on a single series of the number of ill individuals without taking into account the epidemiological features of infectious diseases.
6) According to claim1,2,3,4# the invention is to a we also compare certain specific machine learning techniques. According to experimental data, the suggested strategy produces an MAE of 1.37882 for influenza and 0.69288 for hand, foot, and mouth disease, respectively.