An atopic dermatitis feature prediction method based on enhanced particle swarm optimization

Abstract

The application discloses an atopic dermatitis feature prediction method based on enhanced particle swarm optimization, obtains an individual optimal position based on fitness value comparison; constructs a probability parameter based on a damping factor and a Cauchy random number comparison, and through the introduction of the probability parameter, the population optimal position can accelerate the replacement of individual positions in the early iteration stage, and the population optimal position replaces the individual position less in the late iteration stage to avoid the population individuals from falling into local optimization in the late iteration stage, so that accurate prediction results are obtained; when the result fitness value is obtained, the proportion of the predicted atopic dermatitis features in the total number of features in the feature set is introduced, so that the accurate prediction of atopic dermatitis by fewer predicted features is taken as the standard of the judgment result, and the purpose of efficiently and accurately predicting atopic dermatitis is achieved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of particle swarm optimization, and particularly relates to a characteristic prediction method for atopic dermatitis based on enhanced particle swarm optimization. BACKGROUND

[0002] Atopic dermatitis (AD) is a chronic inflammatory skin disease accompanied by allergic reactions, characterized by pruritus and dry skin in eczematous skin lesions, and is common in children, affecting at least 20% of children worldwide. Considering the starting point of atopic onset accompanied by the development of food allergy, asthma and allergic rhinitis, it is important to distinguish AD and intervention. There are many international diagnostic criteria for AD, such as Hanifin & Rajka diagnostic criteria, Williams diagnostic criteria, International Study of Asthma and Allergies in Childhood (ISAAC) questionnaire, etc., which all depend on the subjective judgment of dermatologists. The standard of William is the main basis for diagnosing AD, including the skin itching condition in the past 12 months and three secondary standards. Due to the lack of experienced doctors, some patients are missed or misdiagnosed. And combining serological results can improve the sensitivity of diagnosis, therefore, combined with serology, comprehensive evaluation of clinical symptoms is gradually valued. However, exploring these problems relies on a large amount of relevant data. In this regard, manual analysis is impractical because it is very time-consuming, inefficient, and prone to errors for a large amount of data, therefore, it is necessary to establish a machine learning model for studying AD.

[0003] Li et al. used three machine learning models to analyze AI-assisted AD diagnosis and classified AD severity through 3D raster scan optoacoustic mesoscopy (RSOM) images to extract features from volumetric vascular structures and clinical information. Jiang et al. developed an accurate and automated method for AD diagnosis based on transcriptomic and microbiome data. By using these data from 161 subjects, including AD patients and healthy controls, they trained a machine learning classifier to predict the risk of AD. The method ultimately proved to be able to accurately distinguish subjects with AD from healthy people according to the relevant data. Holm et al. developed two machine learning models to predict AD and explored the relationship between various immune markers in the serum of AD patients and the severity of AD disease based on clinically obtained biomarkers. Clayton et al. performed a skin biopsy transcriptome analysis of AD. They ultimately revealed the influence of keratinocyte programming on skin inflammation and suggested that interference with the unidirectional immune signal might not be sufficient to address the abnormal cellular immune phenotype of keratinocytes by using co-expression clustering and machine learning tools. However, due to the large difference in the value range of each feature of AD, such as the value range of immunoglobulin E and the content of EO, which can differ by more than 10 times. At the same time, different feature units are different, and the membership relationship is also very different, such as age and dog hair content, which undoubtedly brings not small difficulty to correctly predict AD. Although the existing model can effectively predict AD, the selected features are generally too many, and the accuracy is often not satisfactory. Therefore, we propose a bSRWPSOFKNN model, which can reduce the selected features while ensuring the accuracy through bSRWPSO, and use the FKNN model to predict the selected features, and gradually select important features. SUMMARY

[0004] The application provides an atopic dermatitis feature prediction method based on enhanced particle swarm optimization, which can obtain fewer atopic dermatitis features, so that the atopic dermatitis features obtained can accurately and efficiently predict atopic dermatitis.

[0005] An atopic dermatitis feature prediction method based on enhanced particle swarm optimization, comprising:

[0006] (1) Obtain the feature set of atopic dermatitis patients and the feature category label of atopic dermatitis, and construct a population based on the Sobol sequence with the number of features in the feature set as the dimension;

[0007] (2) comparing the individual position of the current iteration in the population with the individual optimal position of the current iteration in terms of fitness value, taking the position with smaller fitness value as the individual optimal position of the next iteration; constructing the probability parameter of the next iteration, when the attenuation factor is smaller than the Cauchy random number, taking the population optimal position of the current iteration as the probability parameter of the next iteration, when the attenuation factor is larger than the Cauchy random number, taking the individual position of the current iteration as the probability parameter of the next iteration; the attenuation factor decreases constantly with the increase of the iteration number;

[0008] (3) constructing the individual velocity vector of the next iteration based on the individual velocity vector of the current iteration, the difference between the probability parameter of the next iteration and the individual optimal position of the next iteration, and the difference between the probability parameter of the next iteration and the population optimal position of the current iteration; constructing the individual position of the next iteration based on the individual velocity vector of the next iteration and the probability parameter of the next iteration;

[0009] (4) inputting the individual position of the next iteration into the FKNN model to obtain the predicted characteristic category of atopic dermatitis, comparing the predicted characteristic category with the characteristic category label to obtain the error rate value, assigning weights to the error rate value and the proportion of the predicted atopic dermatitis characteristic quantity in the total characteristic quantity in the characteristic set to obtain the result fitness value of the individual position of the next iteration;

[0010] (5) iterating steps (2)-(4) until the iteration number threshold is reached, obtaining a plurality of result fitness values, and taking the predicted characteristic category corresponding to the lowest result fitness value as the final atopic dermatitis prediction characteristic.

[0011] Based on the Sobol sequence, a population is constructed with the number of characteristics in the characteristic set as the dimension, including:

[0012] A matrix is constructed with the number of characteristics in the characteristic set as the dimension and the number of individuals as the row, the random sample point of the Sobol sequence with low variance attribute is S∈[0, 1], each individual in the matrix is initialized based on Sobol to construct the population, the i-th individual X i in the population is:

[0013] X i = lb+S e ·(ub-lb)

[0014] X i ∈[lb, ub]

[0015] Wherein, lb=0, ub=1, i∈[1, N], N is the total number of individuals in the population, S e is the e-th random sample point of the Sobol sequence.

[0016] Each value in the constructed population is represented by 0 or 1, wherein 0 represents that the corresponding feature is not an atopic dermatitis feature, and 1 represents that the corresponding feature is an atopic dermatitis feature.

[0017] The position with a smaller fitness value is taken as the individual optimal position of the next iteration, wherein the i-th individual optimal position pB i (t+1) is:

[0018]

[0019] wherein X i (t) is the individual position of the current iteration, pB i (t) is the individual optimal position of the current iteration, and f(·) is a fitness function.

[0020] The probability parameters of the next iteration are constructed, wherein the j-th probability parameter PL j (t+1) is:

[0021]

[0022] C=tan(pi*(rand-0.5))

[0023] a=1-t / MaxFEs

[0024] wherein gBest(t) is the population optimal position of the current iteration, X i (t) is the individual position of the current iteration, a is a decay factor, C is a Cauchy random number, MaxFEs is the maximum number of iterations, and rand is a random number generated between 0 and 1.

[0025] The g-th individual velocity vector V g (t+1) of the next iteration is:

[0026] V g (t+1)=ω·V g (t)+c1·rand·(pB i (t+1)-PL j (t+1))+c2·rand·(gBest(t)-PL j (t+1))

[0027]

[0028] wherein ω is an adaptive weight value, c1 and c2 are learning factors, rand is a random number generated between 0 and 1, pB i (t+1) is the i-th individual optimal position of the next iteration, PL j(t+1) is the jth probability parameter of the next iteration, gBest(t) is the group optimal position of the current iteration, V max is the boundary value of the individual velocity vector.

[0029] An adaptive weight ω is introduced in the velocity vector to increase the perturbation ability of the velocity vector, and the adaptive weight ω is:

[0030]

[0031] β=1-C·S / MaxFEs

[0032] Wherein, β is the perturbation parameter under the control of C and S, C is a Cauchy random number, S is an adaptive parameter, MaxFEs is the maximum number of iterations, and t is the current number of iterations.

[0033] The adaptive parameter S is:

[0034]

[0035] The ith individual position X i (t+1) of the next iteration is:

[0036] X i (t+1) = PL j (t+1) + V g (t+1)

[0037] Wherein, PL j (t+1) is the jth probability parameter of the next iteration, the gth individual velocity vector V g (t+1) of the next iteration.

[0038] The fitness value Fitness of the individual position of the next iteration is obtained through the individual fitness function based on the error rate value and the proportion of the predicted atopic dermatitis feature quantity in the total feature quantity in the feature set, and the fitness value Fitness is:

[0039]

[0040] Wherein, Error is the error rate value, D is the total number of features in the feature set, R is the predicted atopic dermatitis feature quantity, ε and ∈ are weight parameters, and ε+∈=1.

[0041] Compared with the prior art, the beneficial effects of the present application are:

[0042] The present application can accelerate the replacement of individual positions by introducing a probability parameter to make the optimal position of the population faster in the early iterations, and reduce the number of times of replacing individual positions by the optimal position of the population in the later iterations to avoid the population individuals falling into local optimum in the later iterations, so as to obtain accurate prediction results; when obtaining the fitness value of the result, the proportion of the number of predicted atopic dermatitis characteristics to the total number of characteristics in the characteristic set is introduced, so that the accurate prediction of atopic dermatitis as the standard of the judgment result by fewer predicted characteristics is achieved, so as to achieve the purpose of efficiently and accurately predicting atopic dermatitis. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 A atopic dermatitis feature prediction method based on enhanced particle swarm optimization is provided for the embodiment of the present application.

[0044] Figure 2 The data set provided for the embodiment of the present application is a convergence curve graph of various algorithms. DETAILED DESCRIPTION

[0045] The present application will be explained in detail below in combination with the drawings and specific examples.

[0046] The present application is a atopic dermatitis (AD) prediction and classification method based on enhanced particle swarm optimization, Figure 1 The flowchart of the algorithm is shown in the figure. Feature selection is a discretization problem based on binary. In the final output result, each dimension of each individual represents an attribute of the problem, and each feature has a 0 or 1 label. The final output result is represented by a binary vector, and the length of the vector is based on the number of attributes of the data set. Each value in the vector is represented by 0 or 1, 1 indicating that the attribute is selected, and 0 indicating that the attribute is not selected. For example, the final output result is [0000110001], which indicates that the 5th, 6th and 10th features are selected features, then the selected features are used to judge the category and obtain the result. The specific steps are as follows:

[0047] (1) Obtain the feature set of atopic dermatitis patients as shown in Table 1, and the feature category label of atopic dermatitis, in order to obtain more fair experimental results, 10 times of 10-fold cross-validation is carried out, and ten-fold cross-validation is to divide the training set into 10 sub-samples, and a single sub-sample is reserved as the data for verifying the model, and the other 9 samples are used for training. Cross-validation is repeated 10 times, and each sub-sample is verified once. The results of 10 times are averaged or other combination methods are used to finally obtain a single estimate. The advantage of this method is that the randomly generated sub-samples are repeatedly used for training and verification, and the result is verified once each time. It is preliminarily determined whether each feature in the feature set is an atopic dermatitis feature, and each feature in the feature set is represented by 0 or 1 to construct the first population, wherein 0 represents that the corresponding feature is not an atopic dermatitis feature, and 1 represents that the corresponding feature is an atopic dermatitis feature.

[0048] (2) Construct a population based on Sobol sequence with the number of features in the feature set as the dimension. It can improve the diversity of the population and the ability of the algorithm to traverse the population space through low difference sample points, thereby improving the efficiency of the algorithm to find the global optimal solution. The specific steps are as follows:

[0049] Construct a matrix with the number of features in the feature set as the dimension and the number of individuals as the row. The low-variance random sample points of Sobol sequence are S ∈ [0, 1], and each individual in the matrix is initialized based on Sobol to construct the population. The i-th individual X i in the population is:

[0050] X i = lb + S e · (ub - lb)

[0051] X i ∈ [lb, ub]

[0052] Wherein, lb = 0, ub = 1, i ∈ [1, N], N is the total number of individuals in the population, S e is the e-th random sample point of Sobol sequence.

[0053] (3) Obtain the position of the individual in the next iteration. Each individual in the population is a potential solution to predict atopic dermatitis, which is composed of multiple features. The fitness value of each individual is calculated, and then compared with the fitness value of the previous individual. If it is less than the previous fitness value, it is replaced. In each search process, the optimal position of each individual is recorded by pB, wherein the optimal position of the i-th individual pB i is:

[0054] pB i = (pB i，1 , pB i，2, pB i，3 …pB i，dim )

[0055] wherein dim is dimension.

[0056] The individual position of the current iteration in the population is compared with the individual optimal position of the current iteration in terms of fitness value, and the position with smaller fitness value is taken as the individual optimal position of the next iteration, wherein the i-th individual optimal position pB i (t+1) is:

[0057]

[0058] wherein X i (t) is the individual position of the current iteration, pB i (t) is the individual optimal position of the current iteration, and f(·) is the fitness function.

[0059] The probability parameter of the next iteration is constructed, the population optimal position of the current iteration is taken as the probability parameter of the next iteration when the decay factor is smaller than the Cauchy random number, and the individual position of the current iteration is taken as the probability parameter of the next iteration when the decay factor is larger than the Cauchy random number; the decay factor is constantly reduced with the increase of the iteration number, wherein the j-th probability parameter PL j (t+1) is:

[0060]

[0061] C=tan(π·(rand-0.5))

[0062] a=1-t / MaxFEs

[0063] wherein gBest(t) is the population optimal position of the current iteration, X i (t) is the individual position of the current iteration, a is the decay factor, C is the Cauchy random number, MaxFEs is the maximum iteration number, and rand is a random number generated between 0 and 1. The maximum iteration number of the application is 50, and it can be seen that the probability of algorithm replacement is smaller and smaller in the later period, which can accelerate the convergence of the algorithm in the early period and ensure the diversity of individuals in the later period of the algorithm, avoiding local optimum.

[0064] It can be seen that the random replacement is not simply giving the position of the excellent individual to each individual, but linearly changing through the Cauchy random number, which can approach the target while avoiding falling into local optimum.

[0065] The application provides adjustment of the value of adaptive weight ω. The conventional particle swarm algorithm is prone to local optimum and lacks the ability to jump out of local optimum in the updating process. In order to make up for this deficiency, the application introduces adaptive weight ω in the velocity vector of the conventional particle swarm. The purpose is to increase the perturbation ability of the velocity vector to improve the diversity of individuals in the population, which is conducive to better exploration and utilization of the global optimum by the individuals, and adaptive weight ω is:

[0066]

[0067] β = 1 - C·S / MaxFEs

[0068] Wherein, β is the perturbation parameter under the control of C and S, C is a Cauchy random number, S is an adaptive parameter, MaxFEs is the maximum number of iterations, and t is the current number of iterations. S represents an adaptive parameter with an initial value of 0.01, and the adaptive parameter S is:

[0069]

[0070] Through the adaptive parameter S, the particle swarm algorithm can enhance the convergence speed, and also can adjust its acceleration vector according to the current iteration state, so that the algorithm has a certain ability to escape from local optimum.

[0071] Based on the individual velocity vector of the current iteration, the difference between the probability parameter of the next iteration and the individual optimal position of the next iteration, and the difference between the probability parameter of the next iteration and the population optimal position of the current iteration, the individual velocity vector of the next iteration is constructed, and the velocity vector V g (t+1) of the gth individual of the next iteration is:

[0072] V g (t+1) = ω·V g (t) + c1·rand·(pB i (t+1) - PL j (t+1)) + c2·rand·(gBest(t) - PL j (t+1)

[0073]

[0074] Wherein, ω is the adaptive weight, c1 and c2 are learning factors, rand is a random number generated between 0 and 1, pB i (t+1) is the ith individual optimal position of the next iteration, PL j (t+1) is the jth probability parameter of the next iteration, and gBest(t) is the population optimal position of the current iteration.

[0075] In order to make the individual move within certain limits in order to better search for the value, the particle swarm constraint displacement vector V ∈ [-V max , V max ]. In order to deal with the displacement vector crossing problem, V max is the limit value of the individual velocity vector.

[0076] The individual position of the next iteration is constructed based on the individual velocity vector of the next iteration and the probability parameter of the next iteration. The i-th individual position of the next iteration X i (t+1) is:

[0077] X i (t+1) = PL i (t+1) + V g (t+1)

[0078] Wherein, PL j (t+1) is the j-th probability parameter of the next iteration, and V g (t+1) is the g-th individual velocity vector of the next iteration.

[0079] (4) input the individual position of the next iteration into the FKNN model to obtain the predicted feature category of atopic dermatitis, compare the predicted feature category with the feature category label to obtain an error rate value;

[0080] (5) The accuracy of classification and the number of features are two main targets of feature selection. If a very high classification accuracy is achieved with a small number of features in an experiment, it means that the classification effect is very good. The present application assigns weights to the error rate value and the ratio of the number of predicted atopic dermatitis features to the total number of features in the feature set to obtain the result fitness value of the individual position of the next iteration. The result fitness value Fitness is:

[0081]

[0082] Wherein, Error is the error rate value, D is the total number of features in the feature set, R is the number of predicted atopic dermatitis features, and ε and ∈ are weight parameters, ε+∈=1.

[0083] (6) iterate steps (3)-(5) until the iteration number threshold value of 50 times is reached, obtain a plurality of result fitness values, and the predicted feature category corresponding to the lowest result fitness value is taken as the final atopic dermatitis prediction feature.

[0084] For the 10 datasets in step (1), important features of atopic dermatitis (AD) were predicted using the model in Table 3. These important features were then evaluated using the evaluation criteria in Table 2. The evaluation results are shown in Table 3. This demonstrates that the features predicted by the atopic dermatitis feature prediction method based on enhanced particle swarm optimization provided in this application have high accuracy, sensitivity, Matthews correlation coefficient, and F-metric for AD. Figure 2 As shown, the horizontal axis represents the number of iterations, and the vertical axis represents the fitness value. A smaller fitness value indicates a lower error rate and fewer selected features. In other words, a smaller fitness value means the algorithm can achieve a higher classification accuracy with fewer features, and can correctly predict whether an algorithm is an AD (Alternative Classification) problem. Figure 2 It can be seen that this algorithm has a smaller fitness than other algorithms, meaning it can correctly predict AD with a small number of important features.

[0085] Table 1. Characteristic set of 181 AD patients

[0086]

[0087]

[0088] Table 2. Detailed Explanation of Evaluation Criteria

[0089]

[0090] In Table 2, TP is a true positive number, correctly predicting the number of positive cases as positive; FP is a false positive case, where the negative class was incorrectly predicted as the positive class number; FN is a false negative, where the positive class was correctly predicted as the negative class number; TN is a true negative number, the number of negative classes correctly predicted as negative. In other words, TP represents the number of cases correctly classified as AD (Atopic Dermatitis), meaning both the predicted and true values ​​are atopic dermatitis. FP represents the number of cases incorrectly classified as AD, meaning both the predicted and true values ​​are not atopic dermatitis. FN represents the number of cases incorrectly classified as non-AD, meaning both the predicted and true values ​​are not atopic dermatitis. TN represents the number of cases correctly classified as non-AD, meaning both the predicted and true values ​​are not atopic dermatitis.

[0091] Table 3. Rankings of Friedman's Test on AD

[0092]

[0093] To help AD personnel better and more effectively explore the factors and pathogenic characteristics affecting AD, we improved the particle swarm optimization algorithm to predict AD. First, to overcome the shortcomings of slow convergence speed and poor convergence accuracy of the particle swarm algorithm, we proposed an improved particle swarm optimization algorithm combined with Sobol sequence population initialization, random replacement strategy and adaptive weight strategy, called SRWPSO. In SRWPSO, first, the Sobol sequence is used to initialize the particle swarm due to its low difference sequence distribution and uniformity, which increases the diversity of the initial population and enhances the algorithm's ergodicity in the population space, making it easier for the particle swarm to find the optimal individual position at the beginning. Second, the random replacement strategy and adaptive weight strategy are introduced to make the original particle swarm algorithm overcome the shortcomings of poor convergence ability and local optimal. To apply SRWPSO to medical feature prediction, we proposed a binary version of SRWPSO, named bSRWPSO. Then, we combined bSRWPSO with fuzzy k-nearest neighbor (FKNN) to propose a new and effective classification prediction model, bSRWPSO-FKNN model. Finally, based on bSRWPSO-FKNN and medical datasets, through 10 times 10-fold cross-validation experiments, we extracted key features affecting AD, including the percentage of lymphocytes (LY) in white blood cells, cat dander, milk, dermatophagoides, chicken feather, pigweed, cod, and immunoglobulin E content. In the context of clinical medical practice, the correctness and effectiveness of the experimental results were verified.

Claims

1. An atopic dermatitis feature prediction method based on enhanced particle swarm optimization, characterized by, The method comprises the following steps: (1) obtaining a feature set of atopic dermatitis patients and a feature category label of atopic dermatitis, and constructing a population based on Sobol sequence with the number of features in the feature set as the dimension; (2) comparing the fitness values of the current iteration individual position and the current iteration individual optimal position in the population, and taking the position with the smaller fitness value as the individual optimal position for the next iteration; constructing the probability parameter for the next iteration, when the decay factor is less than the Cauchy random number, taking the population optimal position of the current iteration as the probability parameter for the next iteration, and when the decay factor is greater than the Cauchy random number, taking the individual position of the current iteration as the probability parameter for the next iteration; the decay factor decreases with the increase of the number of iterations; (3) constructing the individual velocity vector for the next iteration based on the individual velocity vector of the current iteration, the difference between the probability parameter for the next iteration and the individual optimal position for the next iteration, and the difference between the probability parameter for the next iteration and the population optimal position of the current iteration, and constructing the individual position for the next iteration based on the individual velocity vector for the next iteration and the probability parameter for the next iteration; (4) inputting the individual position for the next iteration into the FKNN model to obtain the predicted feature category of atopic dermatitis, comparing the predicted feature category with the feature category label to obtain the error rate value, assigning weights to the error rate value and the proportion of the number of predicted atopic dermatitis features to the total number of features in the feature set to obtain the result fitness value of the individual position for the next iteration; (5) iterating steps (2)-(4) until the threshold number of iterations is reached, obtaining a plurality of result fitness values, and taking the predicted feature category corresponding to the lowest result fitness value as the final atopic dermatitis prediction feature.

2. The method of predicting atopic dermatitis characteristics based on enhanced particle swarm optimization according to claim 1, characterized in that, The population is constructed based on Sobol sequence with the number of features in the feature set as the dimension, comprising: A matrix is constructed with the number of features in the feature set as the dimension and the number of individuals as the row. The random sample points of the Sobol sequence with the low variance property are S ∈ [0, 1]. Each individual in the matrix is initialized based on Sobol to construct a population. The i-th individual X i is: X i = lb + S e · (ub - lb) X i ∈[lb,ub] where lb= 0, ub= 1, i∈[1, N], N is the total number of individuals in the population, S e is the e-th random sample point of the Sobol sequence.

3. The method of predicting atopic dermatitis characteristics based on enhanced particle swarm optimization according to claim 1, characterized in that, Each value in the constructed population is represented by 0 or 1, wherein 0 indicates that the corresponding feature is not an atopic dermatitis feature, and 1 indicates that the corresponding feature is an atopic dermatitis feature.

4. The method of predicting atopic dermatitis characteristics based on enhanced particle swarm optimization according to claim 1, characterized in that, The position with a smaller fitness value is taken as the individual optimal position of the next iteration, wherein the ith individual optimal position pB i (t+1) is: where X i (t) is the individual position of the current iteration, pB i (t) is the individual optimal position of the current iteration, f(·) is the fitness function.

5. The method of predicting atopic dermatitis characteristics based on enhanced particle swarm optimization according to claim 1, wherein, constructing the probability parameters for the next iteration, wherein the jth probability parameter PL j (t+1) is: C = tan(π·(rand-0.5)) a = 1-t / MaxFEs where gBest(t) is the population optimal position of the current iteration, X i (t) is the individual position of the current iteration, a is the attenuation factor, C is the Cauchy random number, MaxFEs is the maximum number of iterations, and rand is a random number generated between [0, 1].

6. The method of predicting atopic dermatitis characteristics based on enhanced particle swarm optimization according to claim 1, wherein, the gth individual velocity vector V of the next iteration g (t+1) is: V g (t+1) = ω · V g (t) + c1 · rand · (pB i (t+1) - PL j (t+1)) + c2 · rand · (gBest(t) - PL j (t+1)) where v is the adaptive weight, c1 and c2 are learning factors, rand is a random number generated between 0 and 1, and pB i (t+1) is the i-th individual optimal position of the next iteration, PL j (t+1) is the j-th probability parameter of the next iteration, gBest(t) is the population optimal position of the current iteration, V max is the bound value of the individual velocity vector.

7. The method of predicting atopic dermatitis characteristics based on enhanced particle swarm optimization according to claim 6, characterized in that, An adaptive weight ω is introduced in the velocity vector to increase the perturbation ability of the velocity vector, and the adaptive weight ω is: β = 1-C·S / MaxFEs Wherein, β is the perturbation parameter under the control of C and S, C is the Cauchy random number, S is the adaptive parameter, MaxFEs is the maximum number of iterations, and t is the current iteration number.

8. The method of predicting atopic dermatitis characteristics based on enhanced particle swarm optimization according to claim 7, characterized in that, The adaptive parameter S is:

9. The method of predicting atopic dermatitis characteristics based on enhanced particle swarm optimization according to claim 1, wherein, The i-th individual position X of the next iteration i (t+1) is: X i (t+1) = PL j (t+1) + V g (t+1) where PL j (t+1) is the jth probability parameter of the next iteration, the gth individual velocity vector V g (t+1) of the next iteration.

10. The method of predicting atopic dermatitis characteristics based on enhanced particle swarm optimization according to claim 1, wherein, The result fitness value Fitness of the individual position for the next iteration is obtained by the individual fitness function based on the error rate value and the proportion of the number of predicted atopic dermatitis features to the total number of features in the feature set: Wherein, Error is the error rate value, D is the total number of features in the feature set, R is the number of predicted atopic dermatitis features, and ε and ∈ are weight parameters, ε+∈=1.

Images