Pu'er tea EGCG content prediction method and system based on machine learning

Abstract

The invention provides a Pu'er tea EGCG content prediction method and system based on machine learning in the technical field of EGCG content prediction, and the method comprises the following steps: S1, carrying out soil sampling on a Pu'er tea planting field to obtain a plurality of soil samples, analyzing 21 soil components of each soil sample, and constructing a data set; s2, carrying out regression analysis through an LASSO regression model, and screening modeling factors from the soil components; s3, performing multi-factor regression analysis on each modeling factor through a COX regression model; s4, creating a column graph model based on each modeling factor; s5, performing performance evaluation and stability evaluation on the column graph model through the ROC curve and the calibration curve; and step S6, obtaining to-be-predicted soil data through the visual interface, inputting the soil data into the line graph model in real time, and outputting a prediction result of the EGCG content of the Pu'er tea. The method has the advantage that the prediction accuracy of the EGCG content of the Pu'er tea is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of EGCG content prediction, and in particular to a method and system for predicting the EGCG content of Pu'er tea based on machine learning. Background Art

[0002] Tea was originally a medicinal product, and later gradually became one of the three most popular beverages in the world. Pu'er tea has attracted much attention for its antibacterial, hypoglycemic, hypolipidemic and other health benefits. EGCG (Epigallocatechin gallate) in Pu'er tea is the main substance of bitterness and astringency. With the help of high-tech achievements, the use of tea has been continuously expanded. EGCG has become an important raw material in the food, daily chemical and pharmaceutical industries. EGCG has a variety of pharmacological effects. In addition to playing an important role in anti-cancer, it also plays an important role in anti-inflammatory, antibacterial, and anti-infection. As a result, there is a need to predict the EGCG content in Pu'er tea, that is, to predict the EGCG content based on the planting environment (soil) of tea trees, so as to optimize the planting environment and continuously increase the EGCG content.

[0003] However, due to the diversity and variability of the planting environment, the accuracy of the traditional method of predicting EGCG content based on planting experience is often unsatisfactory. Therefore, how to provide a Pu'er tea EGCG content prediction method and system based on machine learning to improve the accuracy of Pu'er tea EGCG content prediction has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for predicting the EGCG content of Pu'er tea based on machine learning, so as to improve the accuracy of predicting the EGCG content of Pu'er tea.

[0005] In a first aspect, the present invention provides a method for predicting the EGCG content of Pu'er tea based on machine learning, comprising the following steps:

[0006] Step S1, sampling the soil of a Pu'er tea planting area to obtain a plurality of soil samples, analyzing 21 soil components of each soil sample respectively, annotating each soil sample to construct a data set, and performing a sample expansion operation on the data set;

[0007] Step S2, performing regression analysis on the data set by using a LASSO regression model, and screening out modeling factors from each of the soil components;

[0008] Step S3, performing multi-factor regression analysis on each of the modeling factors using a COX regression model;

[0009] Step S4, creating a nomogram model for predicting EGCG content based on each of the modeling factors;

[0010] Step S5, respectively performing performance evaluation and stability evaluation on the nomogram model through ROC curve and calibration curve, and deploying the nomogram model that passes the evaluation;

[0011] Step S6, obtaining soil data to be predicted through a visualization interface, inputting the soil data into a nomogram model in real time, and outputting a prediction result of the EGCG content of Pu'er tea, wherein the visualization interface displays the prediction result of the EGCG content of Pu'er tea in the form of a chart.

[0012] Furthermore, the step S1 is specifically as follows:

[0013] For the sunny and shady slopes of the Pu'er tea planting site, soil sampling was carried out at the top, middle and bottom of the slope, 4cm-24cm below the surface soil, to obtain several soil samples, and the sampling altitude was recorded;

[0014] The 21 soil components of each soil sample, including arsenic content, chromium content, lead content, nickel content, mercury content, effective cadmium content, effective chromium content, effective nickel content, pH value, zinc content, copper content, organic matter content, nitrogen content, phosphorus content, potassium content, effective potassium content, effective phosphorus content, alkaline nitrogen content, magnesium content, fluoride content and cation content, were analyzed respectively, the EGCG content of Pu'er tea corresponding to each soil sample was marked, the EGCG content was associated with the corresponding soil component, a data set was constructed based on each soil component and EGCG content, and sample expansion operation was performed on the data set by Bootstrap method.

[0015] Furthermore, the step S2 is specifically as follows:

[0016] The data set was subjected to regression analysis by a LASSO regression model, and modeling factors significantly correlated with EGCG content were screened out from each of the soil components;

[0017] The loss function of the LASSO regression model is:

[0018]

[0019] Among them, J(θ) represents the loss value of the loss function; θ represents the sample number variable; m represents the number of samples; h θ represents the actual value of the i-th sample; x (i) represents the i-th sample; y (i) represents the predicted value of the i-th sample; λ represents the regularization parameter; n represents the number of features; θ j Represents the regression coefficient of the LASSO regression model.

[0020] Furthermore, the step S3 is specifically as follows:

[0021] A multi-factor regression analysis of the correlation between each modeling factor and the EGCG content was performed using the COX regression model, and each modeling factor after the multi-factor regression analysis was further screened using the Akaike criterion;

[0022] The formula of the COX regression model is:

[0023] h(t)=h0(t)exp(β1x1+...+β j x j );

[0024] Wherein, h(t) represents the risk function of EGCG content; h0(t) represents the intercept of COX regression equation; x j represents the jth modeling factor; β j represents the jth regression coefficient.

[0025] Furthermore, the step S5 is specifically as follows:

[0026] The performance of the nomogram model is evaluated by the AUC value of the ROC curve, and the nomogram model is fitted by a calibration curve to perform stability evaluation. The nomogram model that passes the evaluation is then deployed, and an API interface called by the nomogram model is set.

[0027] In a second aspect, the present invention provides a Pu'er tea EGCG content prediction system based on machine learning, comprising the following modules:

[0028] A data set construction module is used to obtain several soil samples by sampling the soil at the Pu'er tea planting site, analyze the 21 soil components of each soil sample, annotate each soil sample to construct a data set, and perform sample expansion operations on the data set;

[0029] A modeling factor screening module, used for performing regression analysis on the data set through a LASSO regression model to screen out modeling factors from each of the soil components;

[0030] A multi-factor regression analysis module, used to perform multi-factor regression analysis on each of the modeling factors through a COX regression model;

[0031] A nomogram model creation module, used to create a nomogram model for predicting EGCG content based on each of the modeling factors;

[0032] A nomogram model evaluation module, used to respectively evaluate the performance and stability of the nomogram model through ROC curve and calibration curve, and deploy the nomogram model that passes the evaluation;

[0033] The EGCG content prediction module is used to obtain soil data to be predicted through a visual interface, input the soil data into a nomogram model in real time, and output the prediction result of the EGCG content of Pu'er tea. The visual interface displays the prediction result of the EGCG content of Pu'er tea in the form of a chart.

[0034] Furthermore, the data set construction module is specifically used for:

[0035] For the sunny and shady slopes of the Pu'er tea planting site, soil sampling was carried out at the top, middle and bottom of the slope, 4cm-24cm below the surface soil, to obtain several soil samples, and the sampling altitude was recorded;

[0036] The 21 soil components of each soil sample, including arsenic content, chromium content, lead content, nickel content, mercury content, effective cadmium content, effective chromium content, effective nickel content, pH value, zinc content, copper content, organic matter content, nitrogen content, phosphorus content, potassium content, effective potassium content, effective phosphorus content, alkaline nitrogen content, magnesium content, fluoride content and cation content, were analyzed respectively, the EGCG content of Pu'er tea corresponding to each soil sample was marked, the EGCG content was associated with the corresponding soil component, a data set was constructed based on each soil component and EGCG content, and sample expansion operation was performed on the data set by Bootstrap method.

[0037] Furthermore, the modeling factor screening module is specifically used for:

[0038] The data set was subjected to regression analysis by a LASSO regression model, and modeling factors significantly correlated with EGCG content were screened out from each of the soil components;

[0039] The loss function of the LASSO regression model is:

[0040]

[0041] Among them, J(θ) represents the loss value of the loss function; θ represents the sample number variable; m represents the number of samples; h θ represents the actual value of the i-th sample; x (i) represents the i-th sample; y (i) represents the predicted value of the i-th sample; λ represents the regularization parameter; n represents the number of features; θ j Represents the regression coefficient of the LASSO regression model.

[0042] Furthermore, the multi-factor regression analysis module is specifically used for:

[0043] A multi-factor regression analysis of the correlation between each modeling factor and the EGCG content was performed using the COX regression model, and each modeling factor after the multi-factor regression analysis was further screened using the Akaike criterion;

[0044] The formula of the COX regression model is:

[0045] h(t)=h0(t)exp(β1x1+...+β j x j );

[0046] Wherein, h(t) represents the risk function of EGCG content; h0(t) represents the intercept of COX regression equation; x j represents the jth modeling factor; β j represents the jth regression coefficient.

[0047] Furthermore, the nomogram model evaluation module is specifically used for:

[0048] The performance of the nomogram model is evaluated by the AUC value of the ROC curve, and the nomogram model is fitted by a calibration curve to perform stability evaluation. The nomogram model that passes the evaluation is then deployed, and an API interface called by the nomogram model is set.

[0049] The advantages of the present invention are:

[0050] 1. Several soil samples were obtained by sampling the soil in the Pu'er tea planting area, and 21 soil components of each soil sample were analyzed respectively. Each soil sample was labeled to construct a data set, and sample expansion was performed on the data set. Then, the data set was regressed by the LASSO regression model, and modeling factors were screened out from each soil component. Multifactor regression analysis was performed on each modeling factor by the COX regression model. Then, a nomogram model for predicting EGCG content was created based on each modeling factor. The performance and stability of the nomogram model were evaluated by ROC curve and calibration curve respectively. The nomogram model that passed the evaluation was deployed, and finally the soil data to be predicted was obtained through a visual interface. According to the data, the soil data is input into the nomogram model in real time to obtain the prediction result of the EGCG content of Pu'er tea, and the visualization interface displays the prediction result of the EGCG content of Pu'er tea in the form of charts; that is, soil sampling and analysis are carried out in the Pu'er tea planting area to obtain the soil composition of each soil sample, and then the LASSO regression model and COX regression model are used to screen out the soil components most relevant to the EGCG content from various soil components as modeling factors, and a nomogram model is created based on each modeling factor. After the performance and stability evaluation of the nomogram model, the nomogram model is used to predict the EGCG content of Pu'er tea. This method combines machine learning and statistics, and ultimately greatly improves the accuracy of the prediction of the EGCG content of Pu'er tea.

[0051] 2. The EGCG content of Pu'er tea was predicted by the nomogram model. Since the nomogram model is a very intuitive mathematical model, it can combine multiple factors to predict specific results. Compared with other prediction models, the nomogram model usually displays its results in the form of charts, which greatly improves the intuitiveness of the prediction results. In addition, the nomogram model does not require a large amount of training data sets, which greatly improves the construction speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The present invention will be further described below in conjunction with embodiments with reference to the accompanying drawings.

[0053] Figure 1 The present invention is a flow chart of a method for predicting the EGCG content of Pu'er tea based on machine learning.

[0054] Figure 2 It is a structural schematic diagram of a Pu'er tea EGCG content prediction system based on machine learning of the present invention. DETAILED DESCRIPTION

[0055] The technical solution in the embodiments of the present application has the following overall idea: soil sampling and analysis are performed on the Pu'er tea planting site to obtain the soil composition of each soil sample, and then the soil composition most relevant to the EGCG content is screened out from the various soil compositions through the LASSO regression model and the COX regression model as the modeling factor, a nomogram model is created based on the various modeling factors, and after the performance and stability evaluation of the nomogram model, the nomogram model is used to predict the EGCG content of Pu'er tea, combining machine learning and statistics to improve the accuracy of the prediction of the EGCG content of Pu'er tea.

[0056] Please refer to Figure 1 to Figure 2 As shown, a preferred embodiment of the method for predicting the EGCG content of Pu'er tea based on machine learning of the present invention comprises the following steps:

[0057] Step S1, sampling the soil of a Pu'er tea planting area to obtain a plurality of soil samples, analyzing 21 soil components of each soil sample respectively, annotating each soil sample to construct a data set, and performing a sample expansion operation on the data set;

[0058] Step S2, performing regression analysis on the data set through a LASSO regression model, and screening out modeling factors from each of the soil components; by screening each of the soil components, the complexity of the nomogram model is effectively reduced, thereby obtaining better performance parameters and avoiding overfitting;

[0059] Step S3, performing multi-factor regression analysis on each of the modeling factors using a COX regression model;

[0060] Step S4, creating a nomogram model for predicting EGCG content based on each of the modeling factors;

[0061] The nomogram model is based on multi-factor regression analysis, integrating multiple modeling factors and using line segments with scale lines to draw in proportion on the same plane to express the relationship between the modeling factors. By analyzing the size of the regression coefficient in the nomogram model, the influence of each modeling factor on the prediction of EGCG content can be obtained, and the corresponding score is assigned to the value of each dependent variable. The function conversion relationship between the total score and the EGCG content is used to calculate the predicted probability of EGCG content.

[0062] Step S5, respectively performing performance evaluation and stability evaluation on the nomogram model through ROC curve and calibration curve, and deploying the nomogram model that passes the evaluation;

[0063] Step S6, obtaining soil data to be predicted through a visualization interface, inputting the soil data into a nomogram model in real time, and outputting a prediction result of the EGCG content of Pu'er tea, wherein the visualization interface displays the prediction result of the EGCG content of Pu'er tea in the form of a chart.

[0064] The step S1 is specifically as follows:

[0065] For the sunny and shady slopes of the Pu'er tea planting site, soil sampling was carried out at the top, middle and bottom of the slope, 4cm-24cm below the surface soil, to obtain several soil samples, and the sampling altitude was recorded; in specific implementation, each sampling point was sampled 3 times, and each time the sampling was carried out, the consistency of the sampling position, sampling depth and sampling weight must be ensured;

[0066] Analyze the 21 soil components of each soil sample, including arsenic content, chromium content, lead content, nickel content, mercury content, effective cadmium content, effective chromium content, effective nickel content, pH value, zinc content, copper content, organic matter content, nitrogen content, phosphorus content, potassium content, effective potassium content, effective phosphorus content, alkaline nitrogen content, magnesium content, fluoride content and cation content, mark the EGCG content of Pu'er tea corresponding to each soil sample, associate the EGCG content with the corresponding soil component, construct a data set based on each soil component and EGCG content, and perform sample expansion operation on the data set by Bootstrap method. The basic idea of ​​Bootstrap method is to extract several samples from the data set, conduct repeated experiments to construct multiple different data sets, and finally replace the overall distribution with the sample empirical distribution; use random replacement sampling method to perform corresponding operations, extract a certain number of samples from the data set, and then calculate the statistic to be estimated based on the extracted samples, and calculate the variance and distribution based on the calculation results.

[0067] During specific implementation, the contents of copper, zinc, chromium, nickel and lead are determined by flame atomic absorption spectrophotometry, the total phosphorus is determined by alkali fusion-molybdenum antimony counterspectrophotometry, the effective phosphorus is determined by sodium bicarbonate extraction-molybdenum antimony counterspectrophotometry, the effective potassium is determined by combined extraction-colorimetry, the total nitrogen is determined by Kjeldahl method, the fluoride is determined by ion selective electrode method, the cation exchange capacity is determined by hexaamminecobalt trichloride extraction-spectrophotometry, the pH value is determined by potentiometric method, the arsenic and mercury contents are determined by atomic fluorescence method, the organic matter content is determined by potassium dichromate oxidation-spectrophotometry, and the organic carbon content is determined by multiplying the constant 1.724.

[0068] The description table of the soil composition is as follows:

[0069]

[0070]

[0071] The step S2 is specifically as follows:

[0072] The data set was subjected to regression analysis by using a LASSO (Least Absolute Shrinkage and Selection Operator) regression model, and modeling factors significantly correlated with EGCG content were screened out from each of the soil components;

[0073] The loss function of the LASSO regression model is:

[0074]

[0075] Among them, J(θ) represents the loss value of the loss function; θ represents the sample number variable; m represents the number of samples; h θ represents the actual value of the i-th sample; x (i) represents the i-th sample; y (i) represents the predicted value of the i-th sample; λ represents the regularization parameter. The larger the λ value, the fewer variables (modeling factors) are finally screened out by the LASSO regression model; n represents the number of features; θ j Represents the regression coefficient of the LASSO regression model.

[0076] The LASSO regression model is a biased estimation method for processing data with complex collinearity by reducing the variable set and constructing a penalty function to compress the variable coefficients and make some regression coefficients zero, and retains the advantages of subset shrinkage.

[0077] In the coefficient distribution diagram of LASSO regression, as the horizontal axis (penalty coefficient) increases, the vertical axis (variable coefficient) will gradually be affected by the penalty term and compressed until it is finally compressed to 0 and stops changing. In order to reduce the risk of overfitting, the present invention selects the λ value at one standard error level for LASSO regression analysis.

[0078] The step S3 is specifically as follows:

[0079] A multi-factor regression analysis of the correlation between each modeling factor and the EGCG content was performed using the COX regression model, and each modeling factor after the multi-factor regression analysis was further screened using the Akaike criterion;

[0080] The Akaike criterion is a comparative analysis through the AIC (Akaike Information Criterion) value. The AIC value is a standard for measuring the goodness of statistical model fitting. It is based on the concept of entropy and provides a standard for weighing the complexity of the estimated model and the goodness of fitting data. When selecting the best model from a set of available models, the model with the smallest AIC value is usually selected. The calculation formula of the AIC value is:

[0081] AIC = 2k-ln(L);

[0082] Among them, k represents the model parameter, that is, the modeling factor; L represents the likelihood function.

[0083] There are large differences between models, mainly reflected in the likelihood function term. When the likelihood function difference is not significant, the complexity of the model comes into play. When the model complexity increases (k increases), the likelihood function will also increase, and the AIC value will become smaller; when k is too large, the likelihood function growth rate slows down, resulting in an increase in the AIC value. Models that are too complex are prone to overfitting. The AIC value not only improves the model fit, but also introduces a penalty term to minimize the model parameters, which helps reduce the possibility of overfitting.

[0084] The formula of the COX regression model is:

[0085] h(t)=h0(t)exp(β1x1+…+β j x j );

[0086] Wherein, h(t) represents the risk function of EGCG content; h0(t) represents the intercept of COX regression equation; x j represents the jth modeling factor; β j represents the jth regression coefficient.

[0087] The step S5 is specifically as follows:

[0088] The performance of the nomogram model is evaluated by the AUC value of the ROC curve, and the nomogram model is fitted by a calibration curve to evaluate stability. The nomogram model that passes the evaluation is then deployed, and an API interface called by the nomogram model is set.

[0089] The ROC curve is used to evaluate the generalization performance of the model. It was originally used to detect radar signals in wars. It is currently mainly used to evaluate the performance of models generated by machine learning. The ROC curve can be used to comprehensively evaluate the advantages and disadvantages of different models, and the closer the curve is to the upper left, the better the prediction results. The ROC curve divides the final results into positive and negative. When the model accuracy is finally evaluated, the actual situation is divided into True Positive (TP), False Negative (FN), False Positive (FP), and True Negative (TN). When the predicted value is exactly the same as the true value, it is called True Positive or True Negative; when the predicted value is different from the true value, it is called False Negative or False Positive. The ordinate of the ROC curve is True Positive Chance (TPR). The larger the value, the more accurate the prediction. TRP = TP / (TP+FN). The abscissa of the ROC curve is False Positive Chance (FPR). The smaller the value, the more accurate the prediction. FRP = FP / (FP+TN).

[0090] Since the ROC curve cannot quantitatively evaluate the model performance, the model performance is evaluated by calculating the area under the curve (AUC). The AUC value represents the area under the ROC curve. The larger the AUC value, the stronger its classification ability.

[0091] The calibration curve is a tool to evaluate the degree of fit between the model prediction results and the actual situation, and can be used to evaluate the degree of fit between the nomogram and the changes in EGCG content under abiotic stress.

[0092] A preferred embodiment of a Pu'er tea EGCG content prediction system based on machine learning of the present invention comprises the following modules:

[0093] A data set construction module is used to obtain several soil samples by sampling the soil at the Pu'er tea planting site, analyze the 21 soil components of each soil sample, annotate each soil sample to construct a data set, and perform sample expansion operations on the data set;

[0094] A modeling factor screening module is used to perform regression analysis on the data set through a LASSO regression model to screen out modeling factors from each of the soil components; by screening each of the soil components, the complexity of the nomogram model is effectively reduced, thereby obtaining better performance parameters and avoiding overfitting;

[0095] A multi-factor regression analysis module, used to perform multi-factor regression analysis on each of the modeling factors through a COX regression model;

[0096] A nomogram model creation module, used to create a nomogram model for predicting EGCG content based on each of the modeling factors;

[0097] The nomogram model is based on multi-factor regression analysis, integrating multiple modeling factors and using line segments with scale lines to draw in proportion on the same plane to express the relationship between the modeling factors. By analyzing the size of the regression coefficient in the nomogram model, the influence of each modeling factor on the prediction of EGCG content can be obtained, and the corresponding score is assigned to the value of each dependent variable. The function conversion relationship between the total score and the EGCG content is used to calculate the predicted probability of EGCG content.

[0098] A nomogram model evaluation module, used to respectively evaluate the performance and stability of the nomogram model through ROC curve and calibration curve, and deploy the nomogram model that passes the evaluation;

[0099] The EGCG content prediction module is used to obtain soil data to be predicted through a visual interface, input the soil data into a nomogram model in real time, and output the prediction result of the EGCG content of Pu'er tea. The visual interface displays the prediction result of the EGCG content of Pu'er tea in the form of a chart.

[0100] The data set construction module is specifically used for:

[0101] For the sunny and shady slopes of the Pu'er tea planting site, soil sampling was carried out at the top, middle and bottom of the slope, 4cm-24cm below the surface soil, to obtain several soil samples, and the sampling altitude was recorded; in specific implementation, each sampling point was sampled 3 times, and each time the sampling was carried out, the consistency of the sampling position, sampling depth and sampling weight must be ensured;

[0102] Analyze the 21 soil components of each soil sample, including arsenic content, chromium content, lead content, nickel content, mercury content, effective cadmium content, effective chromium content, effective nickel content, pH value, zinc content, copper content, organic matter content, nitrogen content, phosphorus content, potassium content, effective potassium content, effective phosphorus content, alkaline nitrogen content, magnesium content, fluoride content and cation content, mark the EGCG content of Pu'er tea corresponding to each soil sample, associate the EGCG content with the corresponding soil component, construct a data set based on each soil component and EGCG content, and perform sample expansion operation on the data set by Bootstrap method. The basic idea of ​​Bootstrap method is to extract several samples from the data set, conduct repeated experiments to construct multiple different data sets, and finally replace the overall distribution with the sample empirical distribution; use random replacement sampling method to perform corresponding operations, extract a certain number of samples from the data set, and then calculate the statistic to be estimated based on the extracted samples, and calculate the variance and distribution based on the calculation results.

[0103] During specific implementation, the contents of copper, zinc, chromium, nickel and lead are determined by flame atomic absorption spectrophotometry, the total phosphorus is determined by alkali fusion-molybdenum antimony counterspectrophotometry, the effective phosphorus is determined by sodium bicarbonate extraction-molybdenum antimony counterspectrophotometry, the effective potassium is determined by combined extraction-colorimetry, the total nitrogen is determined by Kjeldahl method, the fluoride is determined by ion selective electrode method, the cation exchange capacity is determined by hexaamminecobalt trichloride extraction-spectrophotometry, the pH value is determined by potentiometric method, the arsenic and mercury contents are determined by atomic fluorescence method, the organic matter content is determined by potassium dichromate oxidation-spectrophotometry, and the organic carbon content is determined by multiplying the constant 1.724.

[0104] The description table of the soil composition is as follows:

[0105] English Chinese Arsenic arsenic Chrome chromium Lead lead Nickel nickel Mercury mercury Available Cadmium Effective Cadmium Available Chromium Effective Chromium Available Nickel Effective Nickel PH pH Zn Zinc Cu copper Organic Matter Organic matter N nitrogen P phosphorus K Potassium Available Potassium Available potassium Available Phosphorus Available Phosphorus Alkaline hydrolysis nitrogen Alkaline nitrogen Mg magnesium Fluoride Fluoride Cation cation

[0106] The modeling factor screening module is specifically used for:

[0107] The data set was subjected to regression analysis by using a LASSO (Least Absolute Shrinkage and Selection Operator) regression model, and modeling factors significantly correlated with EGCG content were screened out from each of the soil components;

[0108] The loss function of the LASSO regression model is:

[0109]

[0110] Among them, J(θ) represents the loss value of the loss function; θ represents the sample number variable; m represents the number of samples; h θ represents the actual value of the i-th sample; x (i) represents the i-th sample; y (i) represents the predicted value of the i-th sample; λ represents the regularization parameter. The larger the λ value, the fewer variables (modeling factors) are finally screened out by the LASSO regression model; n represents the number of features; θ j Represents the regression coefficient of the LASSO regression model.

[0111] The LASSO regression model is a biased estimation method for processing data with complex collinearity by reducing the variable set and constructing a penalty function to compress the variable coefficients and make some regression coefficients zero, and retains the advantages of subset shrinkage.

[0112] In the coefficient distribution diagram of LASSO regression, as the horizontal axis (penalty coefficient) increases, the vertical axis (variable coefficient) will gradually be affected by the penalty term and compressed until it is finally compressed to 0 and stops changing. In order to reduce the risk of overfitting, the present invention selects the λ value at one standard error level for LASSO regression analysis.

[0113] The multi-factor regression analysis module is specifically used for:

[0114] A multi-factor regression analysis of the correlation between each modeling factor and the EGCG content was performed using the COX regression model, and each modeling factor after the multi-factor regression analysis was further screened using the Akaike criterion;

[0115] The Akaike criterion is a comparative analysis through the AIC (Akaike Information Criterion) value. The AIC value is a standard for measuring the goodness of statistical model fitting. It is based on the concept of entropy and provides a standard for weighing the complexity of the estimated model and the goodness of fitting data. When selecting the best model from a set of available models, the model with the smallest AIC value is usually selected. The calculation formula of the AIC value is:

[0116] AIC = 2k-ln(L);

[0117] Among them, k represents the model parameter, that is, the modeling factor; L represents the likelihood function.

[0118] There are large differences between models, mainly reflected in the likelihood function term. When the likelihood function difference is not significant, the complexity of the model comes into play. When the model complexity increases (k increases), the likelihood function will also increase, and the AIC value will become smaller; when k is too large, the likelihood function growth rate slows down, resulting in an increase in the AIC value. Models that are too complex are prone to overfitting. The AIC value not only improves the model fit, but also introduces a penalty term to minimize the model parameters, which helps reduce the possibility of overfitting.

[0119] The formula of the COX regression model is:

[0120] h(t)=h0(t)exp(β1x1+…+β j x j );

[0121] Wherein, h(t) represents the risk function of EGCG content; h0(t) represents the intercept of COX regression equation; x j represents the jth modeling factor; β j represents the jth regression coefficient.

[0122] The nomogram model evaluation module is specifically used for:

[0123] The performance of the nomogram model is evaluated by the AUC value of the ROC curve, and the nomogram model is fitted by a calibration curve to evaluate stability. The nomogram model that passes the evaluation is then deployed, and an API interface called by the nomogram model is set.

[0124] The ROC curve is used to evaluate the generalization performance of the model. It was originally used to detect radar signals in wars. It is currently mainly used to evaluate the performance of models generated by machine learning. The ROC curve can be used to comprehensively evaluate the advantages and disadvantages of different models, and the closer the curve is to the upper left, the better the prediction results. The ROC curve divides the final results into positive and negative. When the model accuracy is finally evaluated, the actual situation is divided into True Positive (TP), False Negative (FN), False Positive (FP), and True Negative (TN). When the predicted value is exactly the same as the true value, it is called True Positive or True Negative; when the predicted value is different from the true value, it is called False Negative or False Positive. The ordinate of the ROC curve is True Positive Chance (TPR). The larger the value, the more accurate the prediction. TRP = TP / (TP+FN). The abscissa of the ROC curve is False Positive Chance (FPR). The smaller the value, the more accurate the prediction. FRP = FP / (FP+TN).

[0125] Since the ROC curve cannot quantitatively evaluate the model performance, the model performance is evaluated by calculating the area under the curve (AUC). The AUC value represents the area under the ROC curve. The larger the AUC value, the stronger its classification ability.

[0126] The calibration curve is a tool to evaluate the degree of fit between the model prediction results and the actual situation, and can be used to evaluate the degree of fit between the nomogram and the changes in EGCG content under abiotic stress.

[0127] In summary, the advantages of the present invention are:

[0128] 1. Several soil samples were obtained by sampling the soil in the Pu'er tea planting area, and 21 soil components of each soil sample were analyzed respectively. Each soil sample was labeled to construct a data set, and sample expansion was performed on the data set. Then, the data set was regressed by the LASSO regression model, and modeling factors were screened out from each soil component. Multifactor regression analysis was performed on each modeling factor by the COX regression model. Then, a nomogram model for predicting EGCG content was created based on each modeling factor. The performance and stability of the nomogram model were evaluated by ROC curve and calibration curve respectively. The nomogram model that passed the evaluation was deployed, and finally the soil data to be predicted was obtained through a visual interface. According to the data, the soil data is input into the nomogram model in real time to obtain the prediction result of the EGCG content of Pu'er tea, and the visualization interface displays the prediction result of the EGCG content of Pu'er tea in the form of charts; that is, soil sampling and analysis are carried out in the Pu'er tea planting area to obtain the soil composition of each soil sample, and then the LASSO regression model and COX regression model are used to screen out the soil components most relevant to the EGCG content from various soil components as modeling factors, and a nomogram model is created based on each modeling factor. After the performance and stability evaluation of the nomogram model, the nomogram model is used to predict the EGCG content of Pu'er tea. This method combines machine learning and statistics, and ultimately greatly improves the accuracy of the prediction of the EGCG content of Pu'er tea.

[0129] 2. The EGCG content of Pu'er tea was predicted by the nomogram model. Since the nomogram model is a very intuitive mathematical model, it can combine multiple factors to predict specific results. Compared with other prediction models, the nomogram model usually displays its results in the form of charts, which greatly improves the intuitiveness of the prediction results. In addition, the nomogram model does not require a large amount of training data sets, which greatly improves the construction speed.

[0130] Although the specific implementation modes of the present invention are described above, those skilled in the art should understand that the specific implementation modes described are only illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A method for predicting the EGCG content of Pu'er tea based on machine learning, characterized in that: The steps include: Step S1, sampling the soil of a Pu'er tea planting area to obtain a plurality of soil samples, analyzing 21 soil components of each soil sample respectively, annotating each soil sample to construct a data set, and performing a sample expansion operation on the data set; Step S2, performing regression analysis on the data set by using a LASSO regression model, and screening out modeling factors from each of the soil components; Step S3, performing multi-factor regression analysis on each of the modeling factors using a COX regression model; Step S4, creating a nomogram model for predicting EGCG content based on each of the modeling factors; Step S5, respectively performing performance evaluation and stability evaluation on the nomogram model through ROC curve and calibration curve, and deploying the nomogram model that passes the evaluation; Step S6, obtaining soil data to be predicted through a visualization interface, inputting the soil data into a nomogram model in real time, and outputting a prediction result of the EGCG content of Pu'er tea, wherein the visualization interface displays the prediction result of the EGCG content of Pu'er tea in the form of a chart.

2. A method for predicting the EGCG content of Pu'er tea based on machine learning as claimed in claim 1, characterized in that: The step S1 is specifically as follows: For the sunny and shady slopes of the Pu'er tea planting site, soil sampling was carried out at the top, middle and bottom of the slope, 4cm-24cm below the surface soil, to obtain several soil samples, and the sampling altitude was recorded; The 21 soil components of each soil sample, including arsenic content, chromium content, lead content, nickel content, mercury content, effective cadmium content, effective chromium content, effective nickel content, pH value, zinc content, copper content, organic matter content, nitrogen content, phosphorus content, potassium content, effective potassium content, effective phosphorus content, alkaline nitrogen content, magnesium content, fluoride content and cation content, were analyzed respectively, the EGCG content of Pu'er tea corresponding to each soil sample was marked, the EGCG content was associated with the corresponding soil component, a data set was constructed based on each soil component and EGCG content, and sample expansion operation was performed on the data set by Bootstrap method.

3. A method for predicting the EGCG content of Pu'er tea based on machine learning as claimed in claim 1, characterized in that: The step S2 is specifically as follows: The data set was subjected to regression analysis by a LASSO regression model, and modeling factors significantly correlated with EGCG content were screened out from each of the soil components; The loss function of the LASSO regression model is: Among them, J(θ) represents the loss value of the loss function; θ represents the sample number variable; m represents the number of samples; h θ represents the actual value of the i-th sample; x (i) represents the i-th sample; y (i) represents the predicted value of the i-th sample; λ represents the regularization parameter; n represents the number of features; θ j Represents the regression coefficient of the LASSO regression model.

4. The method for predicting the EGCG content of Pu'er tea based on machine learning as claimed in claim 1, characterized in that: The step S3 is specifically as follows: A multi-factor regression analysis of the correlation between each modeling factor and the EGCG content was performed using the COX regression model, and each modeling factor after the multi-factor regression analysis was further screened using the Akaike criterion; The formula of the COX regression model is: h(t)=h0(t)exp(β1x1+…+β j x j ); Wherein, h(t) represents the risk function of EGCG content; h0(t) represents the intercept of COX regression equation; x j represents the jth modeling factor; β j represents the jth regression coefficient.

5. The method for predicting the EGCG content of Pu'er tea based on machine learning as claimed in claim 1, characterized in that: The step S5 is specifically as follows: The performance of the nomogram model is evaluated by the AUC value of the ROC curve, and the nomogram model is fitted by a calibration curve to perform stability evaluation. The nomogram model that passes the evaluation is then deployed, and an API interface called by the nomogram model is set.

6. A Pu'er tea EGCG content prediction system based on machine learning, characterized in that: Includes the following modules: A data set construction module is used to obtain several soil samples by sampling the soil at the Pu'er tea planting site, analyze the 21 soil components of each soil sample, annotate each soil sample to construct a data set, and perform sample expansion operations on the data set; A modeling factor screening module, used for performing regression analysis on the data set through a LASSO regression model to screen out modeling factors from each of the soil components; A multi-factor regression analysis module, used to perform multi-factor regression analysis on each of the modeling factors through a COX regression model; A nomogram model creation module, used to create a nomogram model for predicting EGCG content based on each of the modeling factors; A nomogram model evaluation module, used to respectively evaluate the performance and stability of the nomogram model through ROC curve and calibration curve, and deploy the nomogram model that passes the evaluation; The EGCG content prediction module is used to obtain soil data to be predicted through a visual interface, input the soil data into a nomogram model in real time, and output the prediction result of the EGCG content of Pu'er tea. The visual interface displays the prediction result of the EGCG content of Pu'er tea in the form of a chart.

7. A Pu'er tea EGCG content prediction system based on machine learning as claimed in claim 6, characterized in that: The data set construction module is specifically used for: For the sunny and shady slopes of the Pu'er tea planting site, soil sampling was carried out at the top, middle and bottom of the slope, 4cm-24cm below the surface soil, to obtain several soil samples, and the sampling altitude was recorded; The 21 soil components of each soil sample, including arsenic content, chromium content, lead content, nickel content, mercury content, effective cadmium content, effective chromium content, effective nickel content, pH value, zinc content, copper content, organic matter content, nitrogen content, phosphorus content, potassium content, effective potassium content, effective phosphorus content, alkaline nitrogen content, magnesium content, fluoride content and cation content, were analyzed respectively, the EGCG content of Pu'er tea corresponding to each soil sample was marked, the EGCG content was associated with the corresponding soil component, a data set was constructed based on each soil component and EGCG content, and sample expansion operation was performed on the data set by Bootstrap method.

8. A Pu'er tea EGCG content prediction system based on machine learning as claimed in claim 6, characterized in that: The modeling factor screening module is specifically used for: The data set was subjected to regression analysis by a LASSO regression model, and modeling factors significantly correlated with EGCG content were screened out from each of the soil components; The loss function of the LASSO regression model is: Among them, J(θ) represents the loss value of the loss function; θ represents the sample number variable; m represents the number of samples; h θ represents the actual value of the i-th sample; x (i) represents the i-th sample; y (i) represents the predicted value of the i-th sample; λ represents the regularization parameter; n represents the number of features; θ j Represents the regression coefficient of the LASSO regression model.

9. A Pu'er tea EGCG content prediction system based on machine learning as claimed in claim 6, characterized in that: The multi-factor regression analysis module is specifically used for: A multi-factor regression analysis of the correlation between each modeling factor and the EGCG content was performed using the COX regression model, and each modeling factor after the multi-factor regression analysis was further screened using the Akaike criterion; The formula of the COX regression model is: h(t)=h0(t)exp(β1x1+...+β j x j ); Wherein, h(t) represents the risk function of EGCG content; h0(t) represents the intercept of COX regression equation; x j represents the jth modeling factor; β j represents the jth regression coefficient.

10. A Pu'er tea EGCG content prediction system based on machine learning as claimed in claim 6, characterized in that: The nomogram model evaluation module is specifically used for: The performance of the nomogram model is evaluated by the AUC value of the ROC curve, and the nomogram model is fitted by a calibration curve to perform stability evaluation. The nomogram model that passes the evaluation is then deployed, and an API interface called by the nomogram model is set.

Images