CN106485086B - Prediction method of human body composition based on AIC and improved entropy weight method - Google Patents

Abstract

Human body composition prediction method based on AIC and improved entropy weight method, including: S1: select the impedance model, collect data and construct the original feature set F of the physiological information sample; S2: add the original feature set F of the physiological information sample, construct the first Characteristic parameters and second characteristic parameters; S3: Use the Akaike information amount criterion, select the AIC stable model; S4: Calculate the value of AIC, select the feature combination with the smallest AIC value, obtain the characteristic parameter matrix, and analyze the effect of each characteristic parameter on the fitting model Influence, modifying the characteristic parameter matrix; S5: Introduce information entropy to obtain a unified matrix, and calculate the entropy value and weight; S6: Solve the characteristic parameter matrix coefficients to obtain a human body composition fitting model. The established human body composition prediction model can improve the prediction accuracy of human body composition, and provide more effective detection means for human body composition research and clinical application.

Description

Prediction method of human body composition based on AIC and improved entropy weight method

technical field

The invention belongs to the field of bioinformatics, in particular to a human body composition prediction method based on AIC and improved entropy weight method.

Background technique

Changes in body composition reflect changes in physical health to a certain extent. Accurate prediction of body composition is of great significance to the regulation of human nutritional status and the prevention of diseases. There are many parameters that affect body composition, mainly including physiological electrical impedance parameters and general physiological characteristic parameters. There are also highly nonlinear and serious correlations between these physiological parameters, and the existing human body composition models are difficult to meet this need.

With the continuous advancement of medical measurement technology, the measurable physiological characteristics have developed on a large scale, and present the characteristics of few samples and high dimensionality, which brings great challenges to the processing and analysis of traditional physiological data, among which redundant features The existence of indirect aggravated the adverse effects, resulting in insufficient prediction of human body composition.

In view of the above problems, it is necessary to propose a new method for predicting human body composition to solve the above problems.

Contents of the invention

Aiming at the deficiencies of the prior art, the present invention proposes a human body composition prediction method based on AIC and improved entropy weight method, and selects an optimal set of characteristic parameters from human physiological parameters, which can effectively reduce the redundancy among characteristic parameters and simplify the fitting model of body composition prediction; secondly, use the improved entropy weight method to solve the unknown coefficients of the prediction model, thereby obtaining the prediction model of human body composition; the human body composition prediction model established in this way can improve the prediction of human body composition Accuracy, providing a more effective detection method for human body composition research and clinical application.

To achieve the above object, the present invention provides a method for predicting human body composition based on AIC and improved entropy weight method, including:

S1: Select an impedance model, collect data and construct an original feature set F of physiological information samples;

S2: Add the original feature set F of the physiological information sample, and construct the first feature parameter and the second feature parameter;

S3: Use the Akaike information criterion and select the AIC stability model;

S4: Calculate the value of AIC, select the feature combination with the smallest AIC value, obtain the feature parameter matrix, analyze the influence of each feature parameter on the fitting model, and correct the feature parameter matrix;

S5: Introduce information entropy and calculate the unified matrix;

S6: Solve the characteristic parameter matrix coefficients to obtain a fitting model of human body composition.

Further, the five-segment impedance value, gender, age, height, weight, and race are the first features; the combination of the square, reciprocal and product of the first feature is the second feature; the original feature set F is composed of the first feature and the second feature Composed together.

Further, the AIC of the Akaike information content criterion is: AIC=2k-ln(L), k is the number of parameters, and L is the likelihood function.

Further, the selected AIC stability model is: AIC _H =logσ ² +(m/n)logn, where σ ² is the variance of the model, m is the highest parameter of the model, and n is the number of parameters.

Further, calculate the value of AIC and select the feature combination with the smallest AIC value to obtain the feature parameter matrix: [R ₁ , R ₃ , R ₄ , R ₂ R ₃ , R ₃ R ₅ ,S,A,H,W, R] ^T , and analyze the influence of each feature parameter on the fitting model, modify the feature parameter matrix, and construct the final feature parameter matrix [R ₁ , R ₃ , R ₄ , R ₂ R ₃ , R ₃ R ₅ ,A,H ,W] ^T , where R ₁ ~ R ₅ are impedance values, S is gender, A is age, H is height, W is weight, R is race, and R _i R _j is the product of impedance values.

Furthermore, the values of race R are completely equal, and the final fitting model constructed is:

Male: f＝a ₁ R ₁ +a ₂ R ₃ +a ₃ R ₄ +a ₄ R ₂ R ₃ +a ₅ R ₃ R ₅ +a ₆ A+a ₇ H+a ₈ W+ε ₁

Female: f＝b ₁ R ₁ +b ₂ R ₃ +b ₃ R ₄ +b ₄ R ₂ R ₃ +b ₅ R ₃ R ₅ +b ₆ A+b ₇ H+b ₈ W+ε ₂

In the formula, a ₁ ~ a ₈ , b ₁ ~ b ₈ are regression coefficients, ε ₁ , ε ₁ are errors.

Furthermore, the information entropy calculation formula is: H(R ₁ )=-∫ _x p(x)logp(x)dx.

Further, the solution steps of the human body composition fitting model are:

S51: Assuming that the evaluation event has m objects and n parameters, X _ij is the j-th index under the i-th object, according to the formula or formula Calculate and determine the decision matrix of m rows and n columns, Y={X _ij } _m×n ;

S52: Eliminate the different dimension units of different indicators of the object to form a unified matrix:

S53: Calculate the entropy value In the formula, e _j is the entropy value corresponding to the jth evaluation index; if Y′ _ij =0, then the value of e _j is in [0,1];

S54: Calculate the weight In the formula, w _j represents the weight of the jth index, and n represents the number of indexes.

As a further step, the steps for solving the human body composition fitting model also include:

S55: Calculate the comprehensive weight:

After calculating the entropy weight of each evaluation index, according to the standard grading number formed by sorting the information entropy of each index, the comprehensive weight of the index x is obtained;

The criterion aggregate entropy is:

Since the importance of each evaluation index has been implied in the grading standard, the conventional weight λ _j is determined by the value of the grading standard. The weight calculation formula is as follows:

Where λ _j is the regular weight of the j index, and k is the standard classification number of the information entropy ranking of the parameter index selected by the feature selection algorithm.

As a further step, a new improved entropy weight is obtained by combining the conventional weight λ _j and the objective weight w _j :

Due to the adoption of the above technical scheme, the present invention can achieve the following technical effects: selecting an optimal set of characteristic parameters from the physiological parameters of the human body can effectively reduce the redundancy among characteristic parameters and simplify the fitting model of body composition prediction ;Secondly, the unknown coefficients of the prediction model are solved by using the improved entropy weight method, so as to obtain the prediction model of human body composition; the prediction model of human body composition established in this way can improve the prediction accuracy of human body composition, and provide great support for human body composition research and clinical application. provide more effective means of detection.

Description of drawings

Fig. 1 is the flowchart of the human body composition prediction method based on AIC and improved entropy weight method of the present invention;

Fig. 2 is a flow chart of the solution steps of the human body composition fitting model;

Figure 3 shows the distribution of human body composition when age is the influencing factor;

Figure 4 shows the distribution of human body composition when body weight is the influencing factor;

Figure 5 shows the distribution of human body composition when height is the influencing factor;

Figure 6 is a schematic diagram of male body composition prediction results;

Figure 7 is a schematic diagram of the prediction results of female body composition.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

Before detailing the prediction method of human body composition based on AIC and improved entropy weight method, the collection of parameters required by this method and related calculation formulas are introduced first.

First, the original feature set of human physiological parameters is constructed. Use the five-segment body impedance measuring instrument to measure the five-segment impedance value and the general physiological characteristic parameters of the human body, and select the five-segment impedance value, gender, age, height, weight, and race as the first feature. Then, the second feature is generated from the first feature through algebraic transformation, that is, the combination of the square, reciprocal and product of the first feature is used as the second feature. The original feature is composed of the first feature and the second feature, that is, the original feature set F consists of five segments of impedance values R1~R5, the combination of impedance values 1/R _i , R _i R _j (1≤i≤5, 1≤j≤5), sex S, age A, height H, weight W, race Z.

Since the characteristic parameters of the fitting model and the selection of their combinations are large and complex, in order to obtain a model that is as simple and accurate as possible, the present invention uses the Akaike Information Criterion AIC as the judging standard to obtain the best explanatory data and contain the least A model with free parameters.

The defined value of AIC is:

AIC=2k-ln(L)

In the formula, k is the number of independent parameters of the model, and L is the maximum likelihood function of the model.

When you want to choose the best model from a group of available models, it is advisable to choose the model with the smallest AIC. When there is a considerable difference between the two models, this difference appears in the above formula L, and when there is no significant difference in L, k works, so a model with a small number of parameters is a good model.

Considering the complexity of the body composition prediction process, this paper chooses the AIC stable model:

AIC _H ＝logσ ² +(m/n)logn

In the formula, ^σ2 is the variance of the model, m is the highest parameter of the model, and n is the number of parameters.

The results of model selection are shown in Table 1, and they are arranged in ascending order of AIC values. In the formula, No. represents the serial number, and n is the number of variables.

Table 1 Model selection results

Based on the results in the above table, select the feature combination whose sequence number is 1, that is, the minimum AIC value, to obtain the feature parameter matrix:

[R ₁ ,R ₃ ,R ₄ ,R ₂ R ₃ ,R ₃ R ₅ ,S,A,H,W,R] ^T

Since the testers selected in this experiment are all Han nationality, the value of race R is completely equal, so the influence on the fitting model is zero, so race R can be removed from the feature parameter matrix. As shown in Figure 3, Figure 4, and Figure 5, there is a direct correlation between gender, age, height, and weight and the human body prediction model, which can be directly used to construct a fitting model. In addition, it can be seen from the figure that When age, weight, and height are used as influencing factors, gender differences are always obvious, so gender is treated independently to improve the accuracy of the model.

Therefore, the simplified feature parameter matrix is obtained:

[R ₁ ,R ₃ ,R ₄ ,R ₂ R ₃ ,R ₃ R ₅ ,A,H,W] ^T

Using the feature parameter matrix to build a human body composition prediction model, here are two prediction models according to gender:

Male: f＝a ₁ R ₁ +a ₂ R ₃ +a ₃ R ₄ +a ₄ R ₂ R ₃ +a ₅ R ₃ R ₅ +a ₆ A+a ₇ H+a ₈ W+ε ₁

Female: f＝b ₁ R ₁ +b ₂ R ₃ +b ₃ R ₄ +b ₄ R ₂ R ₃ +b ₅ R ₃ R ₅ +b ₆ A+b ₇ H+b ₈ W+ε ₂

In the formula, a ₁ to a ₈ are the unknown coefficients of the male fitting model, ε ₁ is the error; b ₁ to b ₈ are the unknown coefficients of the female fitting model, and ε ₂ is the error.

If X=[x ₁ ,x ₂ ,x ₃ ,x ₄ ,x ₅ ,x ₆ ,x ₇ ,x ₈ ]=[R ₁ ,R ₃ ,R ₄ ,R ₂ R ₃ ,R ₃ R ₅ , A,H,W],

A=[a ₁ ,a ₂ ,a ₃ ,a ₄ ,a ₅ ,a ₆ ,a ₇ ,a ₈ ], B=[b ₁ ,b ₂ ,b ₃ ,b ₄ ,b ₅ ,b ₆ ,b ₇ ,b ₈ ]

Then the fitting model f can be expressed as:

Male: f(x)=AX'+ε

Female: f(x)=BX'+ε

Example 1

This embodiment provides a human body composition prediction method based on AIC and improved entropy weight method, including:

S1: Considering the differences in various parts of the human body, select a five-segment impedance model, collect data and construct an original feature set F of physiological information samples;

S2: Consider a series of other physiological parameters that affect body composition, add the original feature set F of the physiological information sample, and construct the first feature parameter and the second feature parameter;

At present, the five-segment human body impedance model is the most commonly used segmented impedance model, which takes into account the differences of various parts of the human body, and divides the human body into five segments of impedance: right upper limb, left upper limb, trunk, right lower limb, and left lower limb. In addition to considering the five-segment impedance values R ₁ ~ R ₅ , human body composition modeling also considers a series of other physiological parameters that affect body composition, including gender S, age A, height H, weight W, race R, and other factors. These characteristic parameters are divided into the first characteristic parameter and the second characteristic parameter, wherein R ₁ ~ R ₅ , S, A, H, W, R are the first characteristic parameters; The reciprocal 1/R _i and the product R _i R _j (1≤i≤5, 1≤j≤5) are the second feature parameter, which is the candidate feature parameter of the combined body composition prediction model.

S3: Use the Akaike Information Criterion as the judging standard, and choose the AIC stable model;

The AIC of Akaike's information content criterion is: AIC=2k-ln(L), k is the number of parameters, and L is the likelihood function.

The body composition prediction process is complicated, and the AIC stable model is selected:

AIC _H ＝logσ ² +(m/n)logn

In the formula, ^σ2 is the variance of the model, m is the highest parameter of the model, and n is the number of parameters.

S4: Calculate the value of AIC, select the feature combination with the smallest AIC value, and obtain the feature parameter matrix: [R ₁ ,R ₃ ,R ₄ ,R ₂ R ₃ ,R ₃ R ₅ ,S,A,H,W,R] ^T. Calculate the influence of characteristic parameters on the fitting model, modify the characteristic parameter matrix: [R ₁ ,R ₃ ,R ₄ ,R ₂ R ₃ ,R ₃ R ₅ ,A,H,W] ^T ,

Male: f＝a ₁ R ₁ +a ₂ R ₃ +a ₃ R ₄ +a ₄ R ₂ R ₃ +a ₅ R ₃ R ₅ +a ₆ A+a ₇ H+a ₈ W+ε ₁

Female: f＝b ₁ R ₁ +b ₂ R ₃ +b ₃ R ₄ +b ₄ R ₂ R ₃ +b ₅ R ₃ R ₅ +b ₆ A+b ₇ H+b ₈ W+ε ₂

S5: Introduce information entropy, improve the traditional entropy weight method, and calculate the unified matrix; the information entropy calculation formula is: H(R ₁ )=-∫ _x p(x)logp(x)dx;

S6: Solve the characteristic parameter matrix coefficients to obtain a fitting model of human body composition.

Example 2

As a supplement to Example 1, the steps for solving the human body composition fitting model are as follows:

S51: Assuming that the evaluation event has m objects and n parameters, X _ij is the j-th index under the i-th object, then the decision matrix Y={X _ij } _m×n with m rows and n columns is based on the larger the better type Index calculation:

Or the smaller the better index calculation:

S52: Eliminate the different dimension units of different indicators of the object to form a unified matrix: In order to make ln(Y′ _ij ) meaningful, it can generally be assumed that: when Y′ _ij =0, Y′ _ij ln(Y′ _ij )=0. But when Y′ _ij =1, ln(Y′ _ij ) is also equal to 0, which is obviously inconsistent with the reality and contrary to the meaning of entropy. Therefore, Y′ _ij is modified:

S53: Calculate the entropy value In the formula, e _j is the entropy value corresponding to the jth evaluation index; if Y′ _ij =0, then the value of e _j is in [0,1];

S54: Calculate the weight In the formula, w _j represents the weight of the jth index, and n represents the number of indexes;

S55: Calculate the comprehensive weight:

Entropy is a measure of uncertainty, and entropy weight reflects the evaluation role of indicators in objective information, and is an objective weight. First use the idea of entropy weight method to calculate the entropy weight of each evaluation index, and then obtain the comprehensive weight of the index x by sorting the standard grading number formed according to the information entropy of each index;

The criterion aggregate entropy is:

Since the importance of each evaluation index has been implied in the grading standard, the conventional weight λ _j is determined by the value of the grading standard. The weight calculation formula is as follows:

Where λ _j is the regular weight of the j index, and k is the standard classification number of the information entropy ranking of the parameter index selected by the feature selection algorithm.

A new improved entropy weight is obtained by combining the conventional weight λ _j and the objective weight w _j :

Use the constructed human body composition model to predict the body composition of 80 men and 80 women in the sample data, and make a relative error comparison chart between the prediction result and the body fat percentage measurement value of InBody770 in Korea, as can be seen from Figure 6 and Figure 7 The relative error values of both men and women are less than 5%. The results show that the predicted value and measured value of human abdominal fat content based on physiological information entropy and improved entropy weight method show a good correlation, and the prediction has considerable accuracy.

Compared with the prior art, the present invention provides a human body composition prediction method based on AIC and improved entropy weight method, and selects an optimal set of characteristic parameters from human physiological parameters, which can effectively reduce the difference between characteristic parameters. Redundancy, simplify the fitting model of body composition prediction; secondly, use the improved entropy weight method to solve the unknown coefficients of the prediction model, so as to obtain the prediction model of human body composition; the human body composition prediction model established in this way can improve the human body composition. Composition prediction accuracy provides a more effective detection method for human body composition research and clinical application.

The above is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto, and any person familiar with the technical field within the technical scope disclosed in the present invention, according to the technical solution of the present invention Any equivalent replacement or change of the inventive concepts thereof shall fall within the protection scope of the present invention.

Claims (1)

1. The human body composition prediction method based on AIC and improved entropy weight method, is characterized in that, comprises: S1: Select an impedance model, collect data and construct an original feature set F of physiological information samples; S2: Add the original feature set F of the physiological information sample, and construct the first feature parameter and the second feature parameter; S3: Use the Akaike information criterion and select the AIC stability model; S4: Calculate the value of AIC, select the feature combination with the smallest AIC value, obtain the feature parameter matrix, analyze the influence of each feature parameter on the fitting model, and correct the feature parameter matrix; S5: Introduce information entropy, obtain a unified matrix, and calculate entropy and weight; S6: Solve the characteristic parameter matrix coefficients to obtain the body composition fitting model of the human body; The steps to solve the human body composition fitting model are as follows: S51: Assuming that the evaluation event has m objects and n parameters, X _ij is the j-th indicator under the i-th object, according to the formula or formula Calculate and determine the decision matrix Y={X _ij } _m×n of m rows and n columns; S52: Eliminate the different dimension units of different indicators of the object to form a unified matrix: S53: Calculate the entropy value In the formula, e _j is the entropy value corresponding to the jth evaluation indicator; if Y' _ij =0, then the value of e _j is in [0,1]; S54: Calculate the weight In the formula, w _j represents the weight of the jth index, and n represents the number of indexes; S55: Calculating the comprehensive weight: after calculating the entropy weight of each evaluation index, according to the standard grading number formed by sorting the information entropy of each index, the comprehensive weight of the index x is obtained; The criterion aggregate entropy is: Since the importance of each evaluation index has been implied in the grading standard, the conventional weight λ _j is determined by the value of the grading standard. The weight calculation formula is as follows: Where λ _j is the regular weight of the jth indicator, and k is the standard classification number of the information entropy ranking of the parameter indicators selected by the feature selection algorithm; A new improved entropy weight is obtained by combining the conventional weight λ _j and the objective weight w _j : The five-segment impedance value, gender, age, height, weight, and race are the first features; the combination of the square, reciprocal and product of the first feature is the second feature; the original feature set F is composed of the first feature and the second feature; Chichi The information amount criterion AIC is: AIC=2k-ln(L), k is the number of parameters, L is the likelihood function; the selected AIC stability model is: AIC _H ＝logσ ² +(m/n)logn, σ ² is The variance of the model, m is the highest parameter of the model, and n is the number of parameters; calculate the value of AIC and select the feature combination with the smallest AIC value to obtain the feature parameter matrix: [R ₁ , R ₃ , R ₄ , R ₂ R ₃ , R ₃ R ₅ ,S,A,H,W,R] ^T , and analyze the impact of each feature parameter on the fitting model, modify the feature parameter matrix, and construct the final feature parameter matrix [R ₁ ,R ₃ ,R ₄ , R ₂ R ₃ ,R ₃ R ₅ ,A,H,W] ^T , where R ₁ ~ R ₅ is impedance value, S is gender, A is age, H is height, W is weight, R is race, R _i R _j is the product of impedance values; the values of race R are completely equal, and the final fitting model is constructed as follows: Male: f＝a ₁ R ₁ +a ₂ R ₃ +a ₃ R ₄ +a ₄ R ₂ R ₃ +a ₅ R ₃ R ₅ +a ₆ A+a ₇ H+a ₈ W+ε ₁ Female: f＝b ₁ R ₁ +b ₂ R ₃ +b ₃ R ₄ +b ₄ R ₂ R ₃ +b ₅ R ₃ R ₅ +b ₆ A+b ₇ H+b ₈ W+ε ₂ In the formula, a ₁ ~a ₈ , b ₁ ~b ₈ are regression coefficients, ε ₁ , ε ₁ are errors; The information entropy calculation formula is: H(R ₁ )=-∫ _x p(x)logp(x)dx.