Machine learning model generation method, antiseptic effect prediction method, device, and program

Abstract

To predict an antiseptic effect of a composition.SOLUTION: An antiseptic effect prediction device 2 comprises: compounding information acquisition means 22 for acquiring compounding information on a target composition; and prediction means 24 for inputting, to a learned machine learning model 5 generated by machine learning using a plurality of pieces of learning data, which includes compounding information 42 and an antiseptic effect 43 of a composition and which is learning data 4 related to a test result of the antiseptic effect of the composition, the acquired compounding information, to predict an antiseptic effect of the target composition.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to a method, an apparatus and a program for predicting preservative efficacy from formulation information of a composition. [Background technology]

[0002] In order to prevent microbial contamination from occurring until the consumer uses up the product, appropriate formulation design for preservative properties is performed when developing oral compositions such as dentifrices and mouthwashes. As a test method for evaluating the preservative effectiveness of a product, a preservative effectiveness test (also called a preservative effectiveness test or a challenge test) is generally known. For example, Patent Document 1 below discloses an example of a preservative effectiveness test performed on a dentifrice. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2016-79139 A Summary of the Invention [Problem to be solved by the invention]

[0004] In the preservative effectiveness test, it is necessary to actually inoculate a product with several kinds of microorganisms such as bacteria and fungi, and to periodically measure the number of live bacteria after a certain time, and the person who carries out the test needs to have specialized skills and experience. In addition, a test period of about one month is required to carry out the preservative effectiveness test. In recent years, the life cycle of a product has been shortened, and in order to efficiently advance product development, it is required to predict the preservative effectiveness of a product in parallel with product development, for example, at the prototype stage of the product. Patent Document 1 does not mention anything about predicting the results of the preservative effectiveness test.

[0005] The present invention aims to predict the preservative efficacy of a composition. [Means for solving the problem]

[0006] The present invention for solving the above problems includes, for example, the following aspects. (Section 1) A preparation step of preparing a plurality of pieces of learning data related to test results of the preservative efficacy of a composition, the learning data including formulation information and preservative efficacy of the composition; A machine learning model generation method including a learning step of performing machine learning using the training data to generate a trained machine learning model that inputs composition information and outputs preservative efficacy. (Section 2) The plurality of learning data are created for each bacterial species, Item 2. The machine learning model generation method according to item 1, wherein the learning step generates the trained machine learning model for each bacterial species. (Section 3) Item 2. The machine learning model generation method according to item 1, wherein the preservative efficacy includes at least one of information on whether the composition complies with a preservative efficacy standard and the number of remaining bacteria. (Section 4) A formulation information acquisition step of acquiring formulation information of a target composition; A prediction step of predicting the preservative efficacy of the target composition by inputting the acquired blending information into a trained machine learning model generated by the machine learning model generation method according to any one of items 1 to 3; A method for predicting preservative efficacy comprising: (Section 5) a step of acquiring information on a target bacterial species; a learning model identification step of identifying a trained machine learning model to be used for prediction based on the acquired bacterial species information; Further comprising: Item 5. The method for predicting preservative efficacy according to Item 4, wherein the prediction step predicts the preservative efficacy of the target composition for each bacterial species by inputting the obtained blending information into the trained machine learning model specified based on the obtained bacterial species information. (Section 6) A preparation means for preparing a plurality of pieces of learning data relating to test results of the preservative efficacy of a composition, the learning data including composition formulation information and preservative efficacy of the composition; A machine learning model generation device comprising a learning means for performing machine learning using the training data to generate a trained machine learning model in which composition formulation information is input and preservative efficacy is output. (Section 7) A blending information acquisition means for acquiring blending information of a target composition; A prediction means for predicting the preservative effect of the target composition by inputting the acquired blending information into a trained machine learning model generated by the machine learning model generation device according to item 6; A preservative effectiveness prediction device comprising: (Section 8) On the computer, A machine learning program for executing each step of the machine learning model generation method according to any one of items 1 to 3. (Section 9) On the computer, Item 6. A prediction program for executing each step of the method for predicting preservative efficacy according to item 4 or 5. Effect of the Invention

[0007] According to the present invention, the preservative efficacy of a composition can be predicted. [Brief description of the drawings]

[0008] [Figure 1] 1 is a diagram showing a schematic configuration of a preservative effectiveness prediction system according to one embodiment; [Diagram 2] FIG. 2 is a block diagram for explaining functions of a machine learning device according to an embodiment. [Diagram 3] FIG. 2 is a block diagram illustrating a function of a prediction device according to an embodiment. [Figure 4] 1 is a flowchart for explaining a procedure for learning a machine learning model using a machine learning device according to an embodiment. [Diagram 5]1 is an example of a test result of preservative effectiveness prepared as learning data. [Figure 6] 1 is a flowchart for explaining a procedure for predicting preservative efficacy using a prediction device and a trained machine learning model according to an embodiment. [Figure 7] FIG. 1 is a diagram showing the accuracy of a prediction model for performing binary classification created in Example 1. [Figure 8] FIG. 1 is a diagram showing the accuracy of a prediction model for performing numerical prediction created in Example 1. [Figure 9] FIG. 1 shows the results of prediction performed by incorporating prescription information of each sample into the prediction model created in Example 1. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings. In the following description and drawings, the same reference numerals denote the same or similar components, and therefore, redundant description of the same or similar components will be omitted.

[0010] [Anti-corrosive efficacy prediction system] <System Overview> FIG. 1 is a diagram showing a schematic configuration of a preservative effectiveness prediction system according to one embodiment.

[0011] A preservative effectiveness prediction system 100 (hereinafter also simply referred to as system 100) according to one embodiment is a system that acquires formulation information of a composition and predicts the preservative effectiveness of the composition, and is equipped with a machine learning device 1 and a preservative effectiveness prediction device 2 (hereinafter also simply referred to as learning device 1 and prediction device 2, respectively).

[0012] The prediction device 2 predicts the preservative efficacy by inputting the formulation information into a trained machine learning model 5. The machine learning model 5 has been trained in advance by the learning device 1. The learning device 1 trains a machine learning model 5 that predicts the preservative efficacy of a composition based on training data 4 related to test results of the preservative efficacy of the composition, and generates a trained machine learning model 5. The trained machine learning model 5 is provided from the learning device 1 to the prediction device 2. The prediction device 2 predicts the preservative efficacy of the composition by inputting the formulation information of the composition into the input of the trained machine learning model 5 trained by the learning device 1.

[0013] The preservative efficacy predicted for a composition includes a preservative efficacy expressed as a binary classification and a preservative efficacy expressed as a numerical value. The preservative efficacy expressed as a binary classification is information on whether or not a standard for preservative efficacy is met, and is expressed, for example, in a Boolean format. The preservative efficacy expressed as a numerical value is information on the number of remaining bacteria, and is expressed, for example, as a numerical value such as the number of bacteria reduced on the 7th day of the test.

[0014] The learning device 1 and the prediction device 2 are directly or indirectly connected by wire or wirelessly in a manner that allows data transmission and reception via, for example, a network 9. The network 9 can be a public network such as the Internet, or a private network such as an in-house intranet or VPN. In another embodiment, the learning device 1 and the prediction device 2 can be integrated together to be configured as a single standalone device that does not use the network 9.

[0015] <Hardware configuration> The learning device 1 and the prediction device 2 can be configured using, for example, a general-purpose computer, a tablet PC, a smartphone, etc. The learning device 1 and the prediction device 2 can be configured entirely with a general-purpose computer, or part of them can be configured with a tablet PC or a smartphone.

[0016] Each of the learning device 1 and the prediction device 2 includes, as hardware configuration, a processor (not shown) such as a CPU that processes data, a main storage device (not shown) that the processor uses as a work area for data processing, and auxiliary storage devices 19 and 29 that are used for temporarily storing data. The auxiliary storage device 19 appropriately stores data 41, 42, and 43 as learning data 4, a machine learning program P1, and the like. The auxiliary storage device 29 appropriately stores a machine learning model 5, a prediction program P2, and the like.

[0017] In this embodiment, the prediction device 2 includes an input unit 26 and a display unit 27 as hardware configurations. The input unit 26 is a device that accepts input from an operator, and the display unit 27 is a device that displays information to the operator. The input unit 26 can be configured, for example, by an input switch, a pointing device, a keyboard, etc., and the display unit 27 can be configured, for example, by a liquid crystal display, an organic EL display, etc. The input unit 26 and the display unit 27 can also be configured as an integrated touch panel.

[0018] [Machine learning device] FIG. 2 is a block diagram illustrating the functions of the machine learning device according to an embodiment.

[0019] The learning device 1 includes a processing unit 10, which includes, as functional blocks, a preparation unit 11 and a learning unit 12. The preparation unit 11 and the learning unit 12 can be implemented as hardware using an integrated circuit or the like. Alternatively, the preparation unit 11 and the learning unit 12 can be implemented as software by the processor of the learning device 1 reading the machine learning program P1 into the main storage device of the learning device 1 and executing it.

[0020] The preparation unit 11 prepares a plurality of pieces of learning data 4 including composition blending information and preservative efficacy. The learning unit 12 performs machine learning using the learning data 4 to generate a trained machine learning model 5. The learning data 4 is learning data related to test results of the preservative efficacy of the composition. The machine learning model 5 is a machine learning model that inputs the composition blending information and outputs the preservative efficacy. In this embodiment, the plurality of pieces of learning data 4 are created for each bacterial species, and the learning unit 12 performs machine learning for each bacterial species to generate a trained machine learning model 5 for each bacterial species.

[0021] <Learning data> The learning data 4 is a data set used for learning the machine learning model 5, and is a set of data set as explanatory variables in the machine learning model 5 and data set as target variables in the machine learning model 5. The learning data 4 is stored in advance in the auxiliary storage device 19 of the learning device 1.

[0022] In this embodiment, the learning data 4 is learning data related to the test results of the preservative effect of the composition, and includes data 41 related to bacterial species information, data 42 related to the composition's blending information, and data 43 related to the composition's preservative effect. Hereinafter, these data 41, 42, and 43 will also be simply referred to as bacterial species information 41, blending information 42, and preservative effect 43.

[0023] In this embodiment, the learning data 4 includes combination information 42 as an explanatory variable and preservative efficacy 43 as a response variable. In this embodiment, the learning data 4 is created for each bacterial species, and the combination of combination information 42 and preservative efficacy 43, which is a combination of the explanatory variable and the response variable, is created for each bacterial species. The bacterial species information 41 is used during machine learning as information for identifying the machine learning model 5 created for each bacterial species.

[0024] The bacterial species information 41 is the type of microorganism to be inoculated into the sample (composition) in the preservative efficacy test. For example, the bacterial species information 41 includes information on the types of bacteria such as Escherichia coli, Staphylococcus aureus, and Pseudomonas aeruginosa, fungi such as yeast and Aspergillus niger, and various environmental bacteria isolated from the environment.

[0025] The blending information 42 is information about the blending of the sample (composition), and includes information about the names of the blended components and information about the ratios of the blended components. Preferably, the ratios of the blended components are the concentration ratios of the blended components. The concentrations are expressed, for example, as weight percent. It is desirable for the ratios of the blended components for one sample to total 100%.

[0026] In this embodiment, the preservative effectiveness 43 includes information on whether the sample (composition) meets the criteria for preservative effectiveness (e.g., a Boolean binary classification value) and the number of remaining bacteria (e.g., the number of bacteria reduced on the seventh day of the test). The criteria for preservative effectiveness can be the evaluation criteria for preservative effectiveness tests stipulated in, for example, ISO11930 as public standards. In addition to the exemplified ISO11930, a user's own evaluation criteria can also be used.

[0027] In this embodiment, the sample to be subjected to machine learning and prediction is an oral composition such as a dentifrice or mouthwash. The dosage form of the composition is not limited to the dentifrice or mouthwash exemplified in the examples described below, and various compositions such as oral ointment medicines, mouth sprays, throat sprays, lotions, milky lotions, beauty serums, creams, sunscreens, makeup bases, hair growth agents, hair sprays, hair waxes, shampoos, rinses, and hand creams are targeted.

[0028] <Machine learning model> For the machine learning model 5, various known machine learning models classified as supervised learning can be used. An example of a machine learning model classified as supervised learning is shown in Table 1. In classifying machine learning models, supervised learning can be further subdivided into classification type and regression type. In this embodiment, for the machine learning model 5, a gradient boosting tree or a neural network is used.

[0029] [Table 1]

[0030] The learning unit 12 sets the formulation information 42 as an explanatory variable and the preservative effectiveness 43 as a target variable, and learns the machine learning model 5. The trained machine learning model 5 is stored in the auxiliary storage device 19 of the learning device 1. The trained machine learning model 5 is transmitted from the learning device 1 to the prediction device 2 via the network 9 and stored in the auxiliary storage device 29 of the prediction device 2. The prediction device 2 uses the received trained machine learning model 5 to predict the preservative effectiveness of the composition.

[0031] [Prediction device] FIG. 3 is a block diagram illustrating the function of the prediction device according to an embodiment.

[0032] The prediction device 2 includes a processing unit 20, which includes, as functional blocks, a fungus species information acquisition unit 21, a blending information acquisition unit 22, a learning model specification unit 23, and a preservative effect prediction unit 24. These functional blocks can be implemented as hardware using an integrated circuit or the like. Alternatively, these functional blocks can be implemented as software by the processor of the prediction device 2 reading a prediction program into the main memory device of the prediction device 2 and executing it.

[0033] The bacterial species information acquisition unit 21 acquires target bacterial species information, for example, from an operator, and the blending information acquisition unit 22 acquires target blending information, for example, from an operator. These pieces of information may be acquired via the input unit 26, or may be input from another computer device via the network 9. The target bacterial species information is input by selecting from a pull-down menu displayed on the display unit 27, for example.

[0034] The learning model identifying unit 23 identifies a trained machine learning model to be used for prediction based on the acquired bacterial species information. In this embodiment, a trained machine learning model 5 is generated for each bacterial species, and the learning model identifying unit 23 identifies the learning model 5 corresponding to the acquired bacterial species information as the trained machine learning model to be used for prediction.

[0035] The preservative efficacy prediction unit 24 predicts the preservative efficacy of the target composition by inputting the obtained blending information into a trained machine learning model 5 specified based on the obtained fungal species information. The prediction result 6 of the preservative efficacy is recorded in, for example, an auxiliary storage device 29, or displayed on, for example, a display unit 27.

[0036] [Machine learning procedure] FIG. 4 is a flowchart illustrating a procedure for learning a machine learning model using a machine learning device according to an embodiment.

[0037] Machine learning is performed using a learning device 1. The machine learning procedure includes step S1 of preparing learning data 4 to be used for learning a machine learning model 5, and step S2 of learning the machine learning model 5 based on the learning data 4. In this embodiment, the learning data 4 is created for each bacterial species, and machine learning is performed for each bacterial species to generate a trained machine learning model 5 for each bacterial species.

[0038] In step S1 (preparation step), the preparation unit 11 prepares data on the test results of the preservative efficacy of the composition as learning data to be used in machine learning. An example of the test results of the preservative efficacy prepared as learning data is shown in FIG. 5. In the figure, each row corresponds to a test result, and the test result includes the prototype lot number of the composition, the actual test result of the preservative efficacy, and the composition blending information. The learning data 4 prepared in step S1 is stored in the auxiliary storage device 19 of the learning device 1 via a network (not shown) or an input device such as a keyboard or mouse (not shown).

[0039] In step S2 (learning step), the learning unit 12 trains a machine learning model 5 that predicts the preservative efficacy of a composition based on the prepared learning data 4. The learning unit 12 trains the machine learning model 5 by setting the blending information 42 and the preservative efficacy 43 contained in the learning data 4 as an explanatory variable and the preservative efficacy 43 as a target variable, respectively. The trained machine learning model 5 is stored in the auxiliary storage device 19.

[0040] The trained machine learning model 5 is transmitted from the learning device 1 to the prediction device 2 via the network 9 and stored in the auxiliary storage device 29 of the prediction device 2. The prediction device 2 uses the received trained machine learning model 5 to predict the preservative efficacy of the composition.

[0041] [Prediction procedure] FIG. 6 is a flowchart illustrating a procedure for predicting preservative efficacy using a prediction device and a trained machine learning model according to an embodiment.

[0042] The prediction of the preservative efficacy is performed using a prediction device 2. The procedure for predicting the preservative efficacy includes step S3 of acquiring bacterial species information, step S4 of acquiring blending information, step S5 of identifying a machine learning model to be used for the prediction, and step S6 of predicting the preservative efficacy. In this embodiment, a trained machine learning model 5 is generated for each bacterial species. Therefore, target bacterial species information is acquired from, for example, an operator, and the trained machine learning model 5 to be used for the prediction is identified based on the acquired bacterial species information.

[0043] In step S3 (bacterial species information acquisition step), the bacterial species information acquisition unit 21 acquires the target bacterial species information, for example, from an operator, and in step S4 (combination information acquisition step), the combination information acquisition unit 22 acquires the target combination information, for example, from the operator.

[0044] In step S5 (learning model identification step), the learning model identification unit 23 identifies a trained machine learning model to be used for prediction based on the bacterial species information acquired by the bacterial species information acquisition unit 21. In this embodiment, a trained machine learning model 5 is generated for each bacterial species, and the learning model 5 corresponding to the acquired bacterial species information is identified as the trained machine learning model to be used for prediction.

[0045] In step S6 (preservative efficacy prediction step), the preservative efficacy prediction unit 24 predicts the preservative efficacy of the target composition by inputting the blending information acquired by the blending information acquisition unit 22 into the trained machine learning model 5 identified by the learning model identification unit 23. The prediction result 6 of the preservative efficacy is recorded in, for example, the auxiliary storage device 29, or displayed on, for example, the display unit 27.

[0046] As described above, the preservative efficacy prediction system according to one embodiment, i.e., the learning device and prediction device constituting the system, can predict the preservative efficacy of a composition. This makes it possible to predict the results of a preservative efficacy test in advance, without actually conducting the test, which requires a test period of about one month.

[0047] In recent years, the life cycle of products has become shorter, and there is a demand for efficient product development. According to the preservative effectiveness prediction system according to one embodiment, for example, it is possible to predict the preservative effectiveness of a product in parallel with product development at the prototype stage, and this makes it possible to further promote the efficiency of product development.

[0048] [Other formats] Although the present invention has been described above with reference to specific embodiments, the present invention is not limited to the above-described embodiments.

[0049] In the above embodiment, the prediction device 2 is realized as an integrated device, but the prediction device 2 does not need to be an integrated device, and the processor, main memory device, auxiliary memory device 29, etc. may be located in separate places and connected to each other so as to be able to communicate with each other via a network.

[0050] In the above embodiment, each functional block of the prediction device 2 is executed by a single processor, but each functional block does not need to be executed by a single processor and may be executed in a distributed manner by multiple processors. The same applies to the learning device 1 as to the prediction device 2.

[0051] A part or all of the functional blocks of the learning device 1 and the prediction device 2 may be cloud-based in a server device (not shown) connected via a network 9.

[0052] In the above embodiment, the learning device 1 and the prediction device 2 are connected to each other so that they can communicate with each other via a network 9, but these learning device 1 and prediction device 2 can also be connected to each other so that they can exchange data via a recording medium such as a DVD-ROM or a memory card.

[0053] The following examples of the present invention will make the features of the present invention clearer. EXAMPLES

[0054] In Example 1, learning data was prepared, machine learning was actually performed, and the effectiveness of the generated machine learning model (hereinafter referred to as a prediction model) was verified.

[0055] <Data collection> The training data used were prescription information (mixture information) registered in the applicant's in-house database and the actual preservative effectiveness test results associated with it. The training data was extracted from the applicant's in-house database for each of the dentifrice and mouthwash under the following conditions. Dentifrice Formulation: Dentifrice Test: Antiseptic efficacy test Mouthwash Form: Liquid toothpaste, mouthwash Test: Antiseptic efficacy test

[0056] <Data preparation> A certain number of the extracted formulas for each of the dentifrice and mouthwash were used for learning. Note that formulas with special dosage forms, such as non-aqueous formulas, were excluded in advance. As described below, some of the extracted formula information that was not used for learning and the actual preservative efficacy test results associated with it were used as data to verify the effectiveness of the prediction model.

[0057] ·Setting explanatory variables The prescription information contains multiple types of ingredients that make up the dentifrice, and these multiple ingredients were used as explanatory variables to predict the dentifrice. Similarly, multiple ingredients were used as explanatory variables to predict the mouthrinse. Note that ingredients that are used very rarely or in very small amounts were excluded from the explanatory variables in advance. Ingredients with almost the same properties were treated as the same explanatory variable (thickeners, polysaccharides, etc.).

[0058] ·Setting the objective variable The objective variables for binary classification and numerical prediction were set as follows: Binary classification: in-criterion / out-of-criterion Numerical prediction: Reduction in number of bacteria (logarithmic value) on the 7th day of the test

[0059] ·Inoculated bacteria Group 1 bacteria (Escherichia coli, Pseudomonas aeruginosa, Staphylococcus aureus) and Group 2 bacteria (Candida albicans, Aspergillus brasiliensis) were used as the basic inoculum. The toothpaste was inoculated with a certain environmentally isolated strain A, and the mouthwash was inoculated with a certain environmentally isolated strain B. The preservative efficacy was evaluated based on Criteria A of ISO11930 Annex B.

[0060] <Applying learning data to predictive analysis tools> Table 2 shows a list of the learning data applied to the predictive analysis tool. The learning data was converted into a CSV file with data items separated by commas and entered into Prediction One (manufactured by Sony Network Communications), a predictive analysis tool using machine learning. Note that the "*" in the table indicates that the text is hidden.

[0061] [Table 2]

[0062] <Creating a predictive model> The items to be predicted were selected in the predictive analysis tool, and a predictive model was created. For binary classification, the following settings were made: Prediction type: Binary classification Predicted value: Within criteria Evaluation data: Automatically extracted from training data, with cross-validation (number of divisions: 2) For numerical prediction, the following settings were made: Forecast type: Numerical forecast Evaluation data: Automatically extracted from training data, with cross-validation (number of divisions: 2)

[0063] The accuracy of the prediction model for binary classification is shown in FIG. 7. The accuracy of the prediction model for numerical prediction is shown in FIG. 8. As shown in FIG. 7, the accuracy of the prediction results was about 80% to about 90%, which was generally high. Therefore, it was confirmed that the accuracy of the prediction model for binary classification was good. As shown in FIG. 8, the coefficient of determination was about 0.64 to 0.8, which also showed a generally high value. The coefficient of determination is often judged to be a good fit when the value exceeds 0.5. Therefore, it was confirmed that the accuracy of the prediction model for numerical prediction was good. From the above, it was confirmed that the prediction accuracy of both the prediction models created for binary classification and numerical prediction was good.

[0064] To supplement understanding, the evaluation items shown in these figures are explained below. AUC (Area Under the Curve) refers to the area under the graph when an ROC (Receiver Operating Characteristic) curve is created. The higher the AUC value, the higher the prediction accuracy. Accuracy refers to the probability that the entire prediction result was correct. Precision refers to the accuracy rate when predicted to be outside the standard. Recall refers to the probability that something that actually is outside the standard was predicted to be outside the standard. F-value refers to the average of Precision and Recall. RMSE (Root Mean Squared Error) refers to the root mean squared error, and in the best case, the value is zero. If there is data that is significantly off the prediction, the value tends to be large. The coefficient of determination is an evaluation value of accuracy based on the average value of squared error. The higher the value, the better the accuracy, with a maximum value of 1.0.

[0065] <Verification of the validity of the prediction model> The validity of the prediction model was evaluated using data that was not used for learning. The following 24 specimens were used for the evaluation data. The specimens were prepared so that the actual preservative efficacy test results for each bacterial group of each formulation included both those that were within the standard and those that were outside the standard. Sample name 001-012 In-house improved toothpaste A Sample name 013~024 In-house improved mouthwash product B

[0066] Predictions were made by inputting the prescription information of each sample into the created binary classification prediction model. The prediction results are shown in Figure 9. Regardless of whether the actual preservative effectiveness test results were within or outside the standards, the prediction results from the prediction model for all 24 samples were consistent with the judgment results from the actual preservative effectiveness test. Therefore, the created prediction model was confirmed to be effective. [Explanation of symbols]

[0067] 1 Machine learning device 2. Preservative efficacy prediction device 4. Training Data 5. Machine learning models 6 Prediction results of preservative effectiveness 9 Network 10 Processing section 11 Preparation Department (preparation means) 12 Learning section (learning means) 19 Auxiliary storage device 20 Processing section 21 Bacterial species information acquisition unit (bacterial species information acquisition means) 22 Mixture information acquisition unit (composition information acquisition means) 23 Learning model identification unit (learning model identification means) 24 Preservative efficacy prediction unit (preservative efficacy prediction means) 26 Input section 27 Display section 29 Auxiliary storage device 41 Bacteria species information data 42 Formulation information data 43 Preservative Efficacy Data 100 Preservative Efficacy Prediction System P1 Machine Learning Program P2 Prediction Program

Claims

1. A preparation step of preparing a plurality of pieces of learning data related to test results of the preservative efficacy of a composition, the learning data including formulation information and preservative efficacy of the composition; A machine learning model generation method including a learning step of performing machine learning using the training data to generate a trained machine learning model that inputs composition information and outputs preservative efficacy.

2. The plurality of learning data are created for each bacterial species, The machine learning model generating method according to claim 1 , wherein the learning step generates the trained machine learning model for each bacterial species.

3. The machine learning model generating method according to claim 1 , wherein the preservative effectiveness includes at least one of information on whether the composition complies with a preservative effectiveness standard and a residual bacterial count.

4. A formulation information acquisition step of acquiring formulation information of a target composition; A prediction step of predicting the preservative effect of the target composition by inputting the obtained blending information into a trained machine learning model generated by the machine learning model generation method according to claim 1; A method for predicting preservative efficacy comprising:

5. a step of acquiring information on a target bacterial species; a learning model identification step of identifying a trained machine learning model to be used for prediction based on the acquired bacterial species information; Further comprising: The preservative efficacy prediction method according to claim 4, wherein the prediction step predicts the preservative efficacy of the target composition for each bacterial species by inputting the acquired formulation information into the trained machine learning model identified based on the acquired bacterial species information.

6. A preparation means for preparing a plurality of pieces of learning data relating to test results of the preservative efficacy of a composition, the learning data including composition formulation information and preservative efficacy of the composition; A machine learning model generation device comprising a learning means for performing machine learning using the training data to generate a trained machine learning model in which composition formulation information is input and preservative efficacy is output.

7. A blending information acquisition means for acquiring blending information of a target composition; A prediction means for predicting the preservative effect of the target composition by inputting the obtained blending information into a trained machine learning model generated by the machine learning model generating device according to claim 6; A preservative efficacy prediction device comprising:

8. On the computer, A machine learning program for executing each step of the machine learning model generation method according to any one of claims 1 to 3.

9. On the computer, A prediction program for executing each step of the method for predicting preservative efficacy according to claim 4 or 5.

Images