CN112381287A - Chinese developmental reading disorder prediction system and prediction method thereof - Google Patents
Chinese developmental reading disorder prediction system and prediction method thereof Download PDFInfo
- Publication number
- CN112381287A CN112381287A CN202011265363.XA CN202011265363A CN112381287A CN 112381287 A CN112381287 A CN 112381287A CN 202011265363 A CN202011265363 A CN 202011265363A CN 112381287 A CN112381287 A CN 112381287A
- Authority
- CN
- China
- Prior art keywords
- layer neuron
- neural network
- module
- prediction
- network model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 208000022372 Reading disease Diseases 0.000 title claims abstract description 59
- 206010013932 dyslexia Diseases 0.000 title claims abstract description 59
- 238000000034 method Methods 0.000 title claims abstract description 31
- 210000002569 neuron Anatomy 0.000 claims abstract description 111
- 238000012360 testing method Methods 0.000 claims abstract description 90
- 238000003062 neural network model Methods 0.000 claims abstract description 77
- 230000001149 cognitive effect Effects 0.000 claims abstract description 8
- 230000005540 biological transmission Effects 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 48
- 238000012549 training Methods 0.000 claims description 36
- 210000000349 chromosome Anatomy 0.000 claims description 18
- 238000004422 calculation algorithm Methods 0.000 claims description 16
- 230000002068 genetic effect Effects 0.000 claims description 16
- 238000012546 transfer Methods 0.000 claims description 14
- 230000035484 reaction time Effects 0.000 claims description 10
- 238000011156 evaluation Methods 0.000 claims description 7
- 108090000623 proteins and genes Proteins 0.000 claims description 6
- 230000035772 mutation Effects 0.000 claims description 4
- 238000012216 screening Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 230000008054 signal transmission Effects 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 abstract description 7
- 238000005516 engineering process Methods 0.000 abstract description 5
- 238000010801 machine learning Methods 0.000 abstract description 2
- 238000013210 evaluation model Methods 0.000 description 8
- 238000007477 logistic regression Methods 0.000 description 5
- 230000007547 defect Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 238000013479 data entry Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 230000004936 stimulating effect Effects 0.000 description 2
- 241000282472 Canis lupus familiaris Species 0.000 description 1
- 241000283707 Capra Species 0.000 description 1
- 235000010627 Phaseolus vulgaris Nutrition 0.000 description 1
- 244000046052 Phaseolus vulgaris Species 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000005034 decoration Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000013551 empirical research Methods 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/451—Execution arrangements for user interfaces
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Abstract
The invention discloses a Chinese developmental reading disorder prediction system and a prediction method thereof, wherein the system comprises a computer and a trained neural network model which is embedded in the computer and runs, the neural network model comprises an input layer neuron module, a hidden layer neuron module and an output layer neuron module, the input layer neuron module and the output layer neuron module are respectively in data transmission with an input module and a prediction module which are arranged on a main interface of the computer, the input layer neuron module comprises a plurality of neuron nodes which are used for inputting tested demographic information, types of participating in a voice consciousness test and reading related cognitive test information, the hidden layer neuron module comprises a plurality of hidden layer neuron nodes, and the output layer neuron module is used for outputting a prediction result of whether the developmental reading disorder exists. The method uses the error reverse transmission neural network technology which can fit any complex and fuzzy data relation by machine learning, and can efficiently and accurately predict the Chinese developmental reading disorder.
Description
Technical Field
The invention relates to the technical field of reading disorder recognition, in particular to a Chinese developmental reading disorder prediction system and a prediction method thereof.
Background
The existing Chinese developmental reading disorder prediction technology is based on the behavior test performance of children, and a traditional Logistic regression model is used for simulation to obtain a final prediction evaluation model (for example, Shu, McBride-Chang, Wu, & Liu, 2006; Li, Shu, McBride-Chang, Liu, & Xue, 2009; Dong, Lihong, Wong, Pan Jing, Zhangping, Raney Fang, 2012; Tong, Tong, & King Yiu, 2018). However, the current predictive assessment model for developmental reading disorder of Chinese children has the following defects:
1) when the Logistic regression method is used for simulating data, the linear, independent, normal distribution and homogeneity of variance of the data need to be met (Lyu & Zhang,2019), but children with reading disorder belong to special groups, and the requirements are difficult to meet in test performance, so that the prediction result of the obtained prediction evaluation model is inaccurate or the prediction deviation is large (Lyu & Zhang, 2019).
2) Developmental reading disorder is the result of mutual influence and interaction of multiple factors (Morris et al, 1998), the complex relationship between the influencing factors is difficult to be accurately characterized by a simple mathematical model of Logistic regression, and the total prediction accuracy of a prediction evaluation model obtained by the Logistic regression method is not high (< 82%).
3) The sample size of simulation data is too small (n is less than 150) when the existing prediction evaluation model is constructed, so that objective rules are difficult to effectively reflect, and the model prediction accuracy is low.
4) The existing prediction and evaluation model is only used as a technical means of empirical research and is not manufactured into a detection system which can be applied to large-scale clinical and scientific research.
Disclosure of Invention
The invention provides a Chinese developmental reading disorder prediction system and a prediction method thereof aiming at the defects that the existing prediction evaluation model data fitting mode of Chinese developmental reading disorder is small in training sample and not applied to reality, and the like, and the prediction system trains a large amount of existing diagnosis data by using an error reverse transfer neural network technology optimized by a genetic algorithm, and finally carries out graphic user interface program design by taking the trained and verified neural network as a core, and is packaged into an evaluation system which can be installed and operated in a Windows 64-bit system and can predict the Chinese developmental reading disorder.
The invention adopts the following technical scheme:
in one aspect, the invention provides a system for predicting Chinese developmental reading disorder, which is characterized in that the system comprises a computer and a trained neural network model embedded in the computer and running, wherein the neural network model comprises an input layer neuron module, a hidden layer neuron module and an output layer neuron module for data signal transmission, and data signals among the input layer neuron module, the hidden layer neuron module and the output layer neuron module are weighted and summed with weight values and biases of each neuron and are sequentially transmitted from left to right through a transfer function among the modules; the input layer neuron module and the output layer neuron module are respectively in data transmission with the input module and the prediction module which are arranged on the main interface of the computer, the input layer neuron module comprises a plurality of neuron nodes which are used for inputting tested demographic information, types of participating in voice consciousness tests and reading related cognitive test information, the hidden layer neuron module comprises a plurality of hidden layer neuron nodes, and the output layer neuron module is used for outputting a prediction result of whether the developing reading disorder exists or not.
The input layer neuron module comprises 14 neuron nodes which are sequentially the tested sex, grade, age, type participating in voice consciousness test, reading fluency, reading accuracy, morpheme consciousness, voice consciousness, picture quick naming, number quick naming, false word accuracy judgment, non-word accuracy judgment, false word reaction time judgment and non-word reaction time achievement judgment; the hidden layer neuron module comprises 12 hidden layer neuron nodes.
The input module comprises an individual prediction module and a group prediction module, wherein the individual prediction module is used for inputting tested demographic information, types of participating in voice consciousness tests and reading related cognitive test information and predicting Chinese developmental reading disorder of input individual data; the group prediction module is used for importing group data and predicting Chinese developmental reading disorder of the imported group data.
The neural network model adopts a train function as a training function, the transfer function adopted by the hidden layer neuron module is an S-type tangent function, and the transfer function adopted by the output layer neuron module is an S-type logarithmic function.
Further, the neural network model is a neural network model which optimizes the initial weight value and the bias through a genetic algorithm, and after the optimized neural network model is trained, the neural network model with the optimal solution is stored and used for predicting the reading disorder of the tested object.
On the other hand, the invention also provides a prediction method based on the Chinese developmental reading disorder prediction system,
step 5, the system takes the normalized input data as an input layer of an input layer neuron module, and performs Chinese developmental reading disorder prediction calculation by calling a sim function and using a trained neural network model;
and 6, directly displaying the prediction result on a main interface of the computer.
The method for constructing the neural network model in the step 5 comprises the following steps:
step 5.1, collecting original screening data of the children with the developmental reading disorder and the normal children, and randomly dividing the original data into a training set and a testing set before training;
step 5.2, using the trainlm function as the training function, setting the expected error to be 1 × 10-5Maximum training epochs is 1000, and learning rate is 0.05;
step 5.3, normalizing the obtained original data;
step 5.4, use formula
H is the number of hidden layer neuron nodes, M is the number of input layer neuron nodes, N is the number of output layer neuron nodes, and alpha is a constant of 1-10; performing point-by-point test on H by using trial-and-error method, fitting each H at least 50 times by using training set data in raw data, and calculating goodness of fit R after each fitting2And root mean square error RMSE, taking R2The average value of the value and the RMSE value is used as an evaluation index of model fitting degree and precision under different node numbers; finally selecting R2H with the highest RMSE and the lowest RMSE is taken as the number of the neuron nodes of the hidden layer;
step 5.5, optimizing the constructed neural network model by using a genetic algorithm, assigning the optimal initial weight value and bias to the constructed neural network model, training the constructed neural network model by using training set data in the original data and storing the final neural network model;
and 5.6, testing the test set in the original data by calling the sim function and using the trained neural network model to obtain a prediction result of the neural network model.
The constructed neural network model is an input layer neuron module containing 14 neuron nodes, a hidden layer neuron module containing 12 hidden layer neuron nodes and an output layer neuron module containing 1 output layer neuron node.
The specific method for optimizing the constructed neural network model by the genetic algorithm in the step 5.5 is as follows:
step 5.5.1, using an ons function to encode the initial weight value and the bias of the constructed 14-12-1 type neural network model as a 'gene' into a 'chromosome', setting the population size to be 50, and using an initializega function to enable the 'chromosome' to form an initial population;
step 5.5.2, putting the weight value and the bias of each chromosome in the established initial population into a neural network model for training and testing, and subtracting an actual result to obtain a test error;
step 5.5.3, taking the reciprocal of the sum of squares of errors of the test result and the actual result as a fitness (fitness) index, comparing the test error obtained in the step 5.5.2 with an expected error, if the obtained test error does not meet the expected error, copying a 'chromosome' with high fitness (setting a selection operator as 0.09), crossing a 'gene' on the 'chromosome' (setting a cross operator as 2) and mutating (setting a mutation operator as [ 21003 ]) to obtain a new population, and repeating the steps of training and testing;
and 5.5.4, if the obtained test error reaches the expected error, decoding the chromosome into an optimal weight value and bias and assigning the optimal weight value and bias to the neural network model.
The ratio of the training set samples to the test set samples contained in the raw data is 7:3, for example, 280 samples may be contained in the training set and 119 samples may be contained in the test set in the raw data.
The technical scheme of the invention has the following advantages:
A. the method uses an error reverse transfer neural network technology which can fit any complex and fuzzy data relation in the field of machine learning, optimizes neural network model parameters by using a genetic algorithm, and finally, takes a neural network model obtained after a large number of data samples are trained and verified as a core to be combined with practical application requirements to manufacture a graphical user interface, so that the method can efficiently and accurately predict the Chinese developmental reading disorder.
B. The method well solves the problem of stuffing shortage existing in the traditional Chinese language children developmental reading disorder prediction model, and improves the overall prediction accuracy to 94.1%, wherein the prediction accuracy for children with reading disorder is 91.4%, and the prediction accuracy for normal children is 96.2%.
Drawings
In order to more clearly illustrate the embodiments of the present invention, the drawings which are needed to be used in the embodiments will be briefly described below, and it is apparent that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained from the drawings without inventive labor to those skilled in the art.
FIG. 1 is a schematic diagram of a neural network model structure in a predictive assessment system according to the present invention;
FIG. 2 is a principal interface of the predictive assessment system provided by the present invention;
FIG. 3 is an individual prediction interface of the predictive assessment system provided by the present invention;
FIG. 4 is a community forecasting interface for the predictive assessment system provided by the present invention;
FIG. 5 is a block diagram of a process for optimizing a neural network model using a genetic algorithm provided by the present invention;
FIG. 6 is a graphical representation comparing the predicted results of the neural network model optimized by the genetic algorithm with the neural network model according to the present invention;
FIG. 7 is a block diagram of a prediction method provided by the present invention.
Detailed Description
The technical solutions of the present invention will be described clearly and completely with reference to the accompanying drawings, and it should be understood that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, and are only for convenience of description and simplicity of description, but do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
In the description of the present invention, it should be noted that, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly, e.g., as meaning either a fixed connection, a removable connection, or an integral connection; the connection can be mechanical connection or electrical connection; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
As shown in fig. 1, the present invention provides a chinese developmental reading disorder prediction system, which includes a computer and a trained neural network model embedded in the computer, where the neural network model includes an input layer neuron module, a hidden layer neuron module, and an output layer neuron module, and data signals between the input layer neuron module, the hidden layer neuron module, and the output layer neuron module are weighted and summed with weight values and biases of each neuron and are sequentially propagated from left to right through transfer functions between the modules; the input layer neuron module and the output layer neuron module are respectively in data transmission with an input module and a prediction module which are arranged on a computer main interface, the input layer neuron module comprises a plurality of neuron nodes which are used for inputting tested demographic information, types of participating in voice consciousness tests and reading related cognitive test information, the hidden layer neuron module comprises a plurality of hidden layer neuron nodes, and the output layer neuron module is used for outputting a prediction result of whether the developing reading disorder exists or not.
The connection of the module and the neuron between the modules in the neural network model is complete, and the neuron of the same module is not connected with each other. During training, two transmission signals exist between the modules, one is a working signal which is transmitted forwards from the input layer neural network model until the output layer neural network module generates an actually output signal; the other is an error signal, i.e. the difference between the actual output value and the expected value, which is transmitted from the output layer neuron module to the input layer neuron module layer by layer. If the neuron module of the output layer can not obtain the expected output, the error signal returns along the original path, the weight value and the bias of each layer of the network are gradually adjusted until the neuron module of the input layer is reached, and the previous forward transmission is repeated.
The input layer neuron module preferably comprises 14 neuron nodes which sequentially comprise tested gender, grade, age, type participating in voice consciousness test, reading fluency, reading accuracy, morpheme consciousness, voice consciousness, quick picture naming, quick number naming, judgment of false word accuracy, judgment of non-word accuracy, judgment of false word reaction time and judgment of non-word reaction time achievement, and each tested original data is obtained through the test; the hidden layer neuron module comprises 12 hidden layer neuron nodes. The neural network model preferably adopts a train function as a training function, the hidden layer neuron module preferably adopts a transfer function of an S-type tangent function, and the output layer neuron module preferably adopts a transfer function of an S-type logarithmic function.
The measurement in the raw data is as follows:
A. reading fluency test: the test contains 160 middle and high frequency Chinese characters, and the children are required to read the Chinese characters quickly and accurately within one minute. And recording the number of the Chinese characters read by the children as a final score.
B. And (3) reading accuracy test: the test contains 172 Chinese characters with high-to-low word frequency, and the Chinese characters recognized by children are required to be accurately read (without time limitation). And recording the number of Chinese characters read by the children as a final result.
C. Morpheme consciousness test: the test contains 20 questions, and requires the children to judge whether the meanings of the same Chinese characters in a pair of double-character words are the same. Such as whether the "letter" words of "envelope" and "trust" mean the same. And recording the total number of questions answered by the children as a final result.
D. And (3) voice consciousness testing:
the test has two forms
Firstly, the oddball paradigm is used to require the children to judge which of the pronunciations of the three Chinese characters presented in sequence is different from the other two pronunciations (10 questions each) in initial consonant, vowel or tone.
Secondly, an oddball paradigm is adopted, and the total number is 30. The test divides the identification of the initial consonant, the final consonant and the tone into 3 sub-tests, and each sub-test has 10 questions. Both tests calculated the total number of questions the child answered as the final performance.
E. Quick automatic naming test:
the test comprises two sub-tests
Picture is named rapidly: the materials are five pictures of flowers, books, dogs, hands and shoes, which are randomly arranged into six rows and five columns, and children are required to read out the names of the pictures in sequence quickly and accurately and record the time.
Quick naming of the numbers: the stimulating material is changed into a few numbers of 2, 4, 6, 7 and 9, and the process is named as the rapid picture. Each sub-test was performed twice and the average was taken as the final performance.
F. Testing consciousness by orthography: the test is a true, false, non-word determination task. The stimulating material includes 40 true characters, 20 false characters (e.g., the same common bean curd), and 20 non-characters (e.g., the different from. The program uses E-prime2.0 presentation, presenting the 500 ms gaze point first, and then presenting the stimulus, asking the child to determine if the stimulus seen is a real existing chinese character. Recording the response and accuracy of the children to the false words and non-words.
The system comprises two types of prediction, namely an individual prediction module and a group prediction module which are respectively displayed on a computer main interface. The individual prediction module is used for inputting tested demographic information, types of participating in voice consciousness tests and reading related cognitive test information and performing Chinese developmental reading disorder prediction on input individual data; the group prediction module is used for importing group data and predicting Chinese developmental reading disorder of the imported group data.
The system further adopts a genetic algorithm to optimize the initial weight value and the bias of the constructed neural network model, trains the optimized neural network model, and then stores the neural network model with the optimal solution for predicting the reading disorder of the tested object.
Aiming at the defects that the existing prediction evaluation model data fitting mode of the Chinese developmental reading disorder is small in training sample and not applied to reality, and the like, the invention trains a large amount of existing diagnostic data by using an error reverse transfer neural network technology optimized by a genetic algorithm, finally, a graphical user interface program is designed by taking the trained and verified neural network as a core, and the graphical user interface program is packaged into an evaluation system which can be installed and operated in a Windows 64-bit system and can predict the Chinese developmental reading disorder.
The core neural network model of the predictive assessment system has a structure of 14-12-1, as shown in FIG. 1. The system comprises 14 input layer neurons (14 variables are sex, grade and age of children, types of children participating in voice consciousness tests, reading fluency, reading accuracy, morpheme consciousness, voice consciousness, picture quick naming, digit quick naming, character correctness judgment, character incorrect judgment, character reaction time judgment and character non-reaction time judgment), 12 hidden layer neurons and 1 output layer neuron (whether the input layer neurons are developmental reading disorder or not).
The main interface of the prediction evaluation system mainly comprises two modules: "individual prediction" and "group prediction", as shown in FIG. 2.
The "individual prediction" module is used to predict a single subject, as shown in fig. 3. In this module, the main test needs to input the demographic information (including name, sex, grade, age) of the tested person, the type of the test taking part in the speech consciousness test and the information of the reading related cognitive test (including reading fluency, reading accuracy, morpheme consciousness, speech consciousness, quick picture naming, quick number naming, judging the correctness of the false word, judging the correctness of the non-word, judging the reaction time of the false word and judging the reaction time of the non-word). After clicking the "predict" button, the system will call the original data set to merge with the input data and normalize. Then, the system takes the normalized input data as an input layer, calls a "sim" function and calculates by using the trained neural network model. Finally, the system pops up a dialog box to prompt whether the tested person suffers from the developmental reading disorder according to the obtained result.
The "community prediction" module is used to predict two or more subjects simultaneously, as shown in FIG. 4. In this module, a main test first needs to enter all tested data information in Excel according to a "data entry example" on a software interface. Then, a data file to be predicted is selected and an output position of a prediction result file is designated. After clicking the "predict" button, the system will call the original dataset to merge with the input dataset and normalize. Then, the system takes the normalized input data set as an input layer, calls a "sim" function and calculates by using the trained neural network model. And finally, judging whether each individual in the group has developmental reading disorder according to the obtained result, and outputting the result to a specified position.
As shown in fig. 7, a specific method for predicting the chinese developmental reading disorder is as follows:
1) data collection: collecting the demographics and test data of children, and deriving 14 data in total, including gender, grade, age, type of taking part in the voice consciousness test, reading fluency, reading accuracy, morpheme consciousness, voice consciousness, quick picture naming, quick number naming, false word accuracy judgment, false word response judgment and non-word response judgment.
2) And (3) data encoding: gender, grade, type of speech awareness test being tested were re-encoded. Gender is re-encoded as "1 as male" and "2 as female". The grades are re-encoded as "3-grade 3", "4-grade 4", "5-grade 5", and "6-grade 6". The type of speech awareness test being tested is re-encoded as "1" on the first exam and "2" on the second exam. The subject age and the reading related skill variables are directly encoded with the raw data.
3) Inputting the tested data to be tested through an input module on a computer main interface;
the basic parameters are set as follows:
A. setting input layer neuron node numbers 14 (including tested gender, grade, age, voice awareness test type and 10 reading related skill achievements); setting the number of hidden layer layers to be 1, and having 12 neuron nodes; the number of output layer neuron nodes 1 (whether or not there is a reading disorder) is set.
B. The hidden layer transfer function is set as an S-type tangent function, and the output layer transfer function is set as an S-type logarithmic function.
4) The system combines the calling original data set with the input data and normalizes the data;
5) the system takes normalized input data as an input layer of an input layer neuron module, calls a sim function and uses a trained neural network model to predict and calculate the Chinese developmental reading disorder;
6) the prediction result is directly displayed on the main interface of the computer.
The method for constructing the neural network model in the step 5 comprises the following steps:
A) collecting original screening data of developing reading disorder children and normal children, and randomly dividing the original data into a training set and a testing set before training, wherein the training set comprises 280 samples (70%), and the testing set comprises 119 samples (30%);
B) using the tranlmm function as the training function, the expected error is set to 1 × 10-5Maximum training epochs is 1000, and learning rate is 0.05;
C) the raw data was normalized.
D) Determining the number of hidden layer neuron nodes: according to empirical formula
Where H is the number of hidden layer neuron nodes (5 (a is 1) to 13(a is 10)), M is the number of input layer neuron nodes, N is the number of output layer neuron nodes, and α is a constant of 1 to 10. The point-by-point test was performed on H using the trial-and-error method, 50 fits were performed on each H using the training set data,calculating goodness of fit R after each fitting2And root mean square error RMSE, eliminating R outside the mean + -2.5 SD2And RMSE, taking the remaining R2And taking the average value of the value and the RMSE value as an evaluation index of the model fitting degree and the precision under the node number. Finally selecting R2The highest and lowest RMSE H are taken as the number of hidden layer neuron nodes (12 is preferred in the present invention).
E) Genetic algorithm optimization: because the basic neural network model is easy to fall into a local optimal solution, the initial weight value and the bias of the constructed neural network model with the structure of 14-12-1 are optimized by using a genetic algorithm (as shown in figure 5), the optimal initial weight value and the bias are assigned to the constructed neural network model, the constructed neural network model is trained by using training set data in original data, and the final neural network model is stored.
Optimizing a neural network model by a genetic algorithm:
coding the initial weight value and the bias of the constructed 14-12-1 type neural network model as a gene into a chromosome by using an ones function, setting the population size to be 50, and forming the chromosome into an initial population by using an initializega function;
secondly, putting the weight value and the bias of each chromosome in the established initial population into a neural network model for training and testing, and subtracting an actual result to obtain a test error;
comparing the test error obtained in the step (II) with an expected error by taking the reciprocal of the sum of squares of errors of the test result and the actual result as an index of fitness (fitness), if the obtained test error does not meet the expected error, copying a 'chromosome' with high fitness (setting a selection operator as 0.09), crossing (setting a cross operator as 2) and mutating (setting a mutation operator as [ 21003 ]) a 'gene' on the 'chromosome' to obtain a new population, and repeating the steps of training and testing; the copying, crossing and mutation processes are realized by calling a GOAT tool box.
And fourthly, if the obtained test error reaches the expected error, decoding the chromosome into the optimal weight value and bias and giving the optimal weight value and bias to the neural network model.
F) And testing the test set in the original data by calling the sim function and using the trained neural network model to obtain the prediction result of the neural network model.
As shown in fig. 2, the chinese developmental reading disorder prediction system is constructed in a computer:
A. three buttons are arranged on the computer main interface, an individual prediction button points to an individual prediction interface, a group prediction button points to a group prediction interface, and an exit system button is used as an exit program.
The 'individual forecast' interface places 20 input text boxes/buttons for entering the data of each item to be tested. Three more function buttons are placed, a "predict" button to predict the individual data entered, a "reset" button to clear all data entered in the text box/button, and a "back" button to the home interface, as shown in fig. 3.
The "group forecast" interface places data entry example pictures and description text boxes, plus five function buttons. A "please select data (Excel)" button is used to import group data, a "please select result save location" button is used to set the save path and file name of the prediction result file, a "prediction" button is used to predict the input group data, a "reset" button is used to clear the input group data and save path, and a "return" button points to the main interface, as shown in fig. 4.
FIG. 6 is a comparison of experimental results of the constructed Basic neural network model (Basic BPNN), the neural network model (GA-BPNN) optimized by the genetic algorithm, and the traditional Logistic regression method, and specific prediction results are shown in the following table.
From the table, the method well solves the problem of stuffing shortage existing in the traditional Chinese language children developmental reading disorder prediction evaluation model, and improves the overall prediction accuracy to 94.1%, wherein the prediction accuracy for reading disorder children is 91.4%, and the prediction accuracy for normal children is 96.2%.
The invention provides a neural network model capable of carrying out prediction evaluation on Chinese developmental reading disorder, and the neural network model is presented in a graphical user interface mode. Therefore, the neural network model, the software structure, the achievable functions and the equivalent changes and decorations included in the present invention are included in the protection scope of the patent, and are not applicable to the prior art.
It should be understood that the above examples are only for clarity of illustration and are not intended to limit the embodiments. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. And obvious variations or modifications therefrom are intended to be within the scope of the invention.
Claims (10)
1. The Chinese developmental reading disorder prediction system is characterized by comprising a computer and a trained neural network model which is embedded in the computer and runs, wherein the neural network model comprises an input layer neuron module, a hidden layer neuron module and an output layer neuron module which are used for data signal transmission, data signals among the input layer neuron module, the hidden layer neuron module and the output layer neuron module are weighted and summed with weight values and biases of all neurons, and the data signals are transmitted from left to right in sequence through transfer functions among the modules; the input layer neuron module and the output layer neuron module are respectively in data transmission with the input module and the prediction module which are arranged on the main interface of the computer, the input layer neuron module comprises a plurality of neuron nodes which are used for inputting tested demographic information, types of participating in voice consciousness tests and reading related cognitive test information, the hidden layer neuron module comprises a plurality of hidden layer neuron nodes, and the output layer neuron module is used for outputting a prediction result of whether the developing reading disorder exists or not.
2. The system for predicting Chinese developmental reading disorder according to claim 1, wherein the input layer neuron module comprises 14 neuron nodes, which are sequentially tested gender, grade, age, type of taking part in voice consciousness test, reading fluency, reading accuracy, morpheme consciousness, voice consciousness, quick picture naming, quick number naming, determining the false word accuracy, determining the non-word accuracy, determining the false word reaction time and determining the non-word reaction time achievement; the hidden layer neuron module comprises 12 hidden layer neuron nodes.
3. The system for predicting Chinese developmental reading disorder according to claim 1, wherein the input module comprises an individual prediction module and a group prediction module, the individual prediction module is used for inputting the tested demographic information, the type of taking part in the speech consciousness test and the reading related cognitive test information, and performing Chinese developmental reading disorder prediction on the input individual data; the group prediction module is used for importing group data and predicting Chinese developmental reading disorder of the imported group data.
4. The system for predicting Chinese developmental reading disorder according to claim 1, wherein a tranlm function is used as the training function in the neural network model, the transfer function used by the hidden layer neuron module is a S-type tannent function, and the transfer function used by the output layer neuron module is a S-type logarithmic function.
5. The Chinese developmental reading disorder prediction system according to any one of claims 1 to 4, wherein the neural network model is a neural network model whose initial weight values and biases are optimized by a genetic algorithm, and after the optimized neural network model is trained, the neural network model with the optimal solution is stored for predicting the reading disorder of the subject.
6. A prediction method based on a Chinese developmental reading disorder prediction system is characterized in that,
step 1, collecting demographics and test data of children;
step 2, recoding the gender, the grade and the type of the tested subject to participate in the voice consciousness test, and directly coding the tested age and the related reading skill variables by using original data;
step 3, inputting the tested data to be tested through an input module on a computer main interface;
step 4, the system combines the calling original data set with the input data and normalizes the data;
step 5, the system takes the normalized input data as an input layer of an input layer neuron module, and performs Chinese developmental reading disorder prediction calculation by calling a sim function and using a trained neural network model;
and 6, directly displaying the prediction result on a main interface of the computer.
7. The prediction method based on the Chinese developmental reading disorder prediction system according to claim 6, wherein the neural network model in the step 5 is constructed by:
step 5.1, collecting original screening data of the children with the developmental reading disorder and the normal children, and randomly dividing the original data into a training set and a testing set before training;
step 5.2, using the trainlm function as the training function, setting the expected error to be 1 × 10-5Maximum training epochs is 1000, and learning rate is 0.05;
step 5.3, normalizing the obtained original data;
step 5.4, use formula
H is the number of hidden layer neuron nodes, M is the number of input layer neuron nodes, N is the number of output layer neuron nodes, and alpha is a constant of 1-10; using trial and error method (three-and-error) performing point-by-point test on H, fitting each H at least 50 times by using training set data in original data, and calculating goodness of fit R after each fitting2And root mean square error RMSE, taking R2The average value of the value and the RMSE value is used as an evaluation index of model fitting degree and precision under different node numbers; finally selecting R2H with the highest RMSE and the lowest RMSE is taken as the number of the neuron nodes of the hidden layer;
step 5.5, optimizing the constructed neural network model by using a genetic algorithm, assigning the optimal initial weight value and bias to the constructed neural network model, training the constructed neural network model by using training set data in the original data and storing the final neural network model;
and 5.6, testing the test set in the original data by calling the sim function and using the trained neural network model to obtain a prediction result of the neural network model.
8. The prediction method based on the Chinese developmental reading disorder prediction system according to claim 7, wherein the neural network models are constructed as an input layer neuron module comprising 14 neuron nodes, a hidden layer neuron module comprising 12 hidden layer neuron nodes, and an output layer neuron module comprising 1 output layer neuron node.
9. The prediction method based on the Chinese developmental reading disorder prediction system according to claim 8, wherein the genetic algorithm in the step 5.5 is used for optimizing the constructed neural network model by the following specific methods:
step 5.5.1, using an ons function to encode the initial weight value and the bias of the constructed 14-12-1 type neural network model as a 'gene' into a 'chromosome', setting the population size to be 50, and using an initializega function to enable the 'chromosome' to form an initial population;
step 5.5.2, putting the weight value and the bias of each chromosome in the established initial population into a neural network model for training and testing, and subtracting an actual result to obtain a test error;
step 5.5.3, taking the reciprocal of the sum of squares of errors of the test result and the actual result as a fitness (fitness) index, comparing the test error obtained in the step 5.5.2 with an expected error, if the obtained test error does not meet the expected error, copying a 'chromosome' with high fitness (setting a selection operator as 0.09), crossing a 'gene' on the 'chromosome' (setting a cross operator as 2) and mutating (setting a mutation operator as [ 21003 ]) to obtain a new population, and repeating the steps of training and testing;
and 5.5.4, if the obtained test error reaches the expected error, decoding the chromosome into an optimal weight value and bias and assigning the optimal weight value and bias to the neural network model.
10. The prediction method based on the Chinese developmental reading disorder prediction system according to claim 7, wherein in the step 5.1, the ratio of the training set samples to the test set samples contained in the original data is 7: 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011265363.XA CN112381287A (en) | 2020-11-13 | 2020-11-13 | Chinese developmental reading disorder prediction system and prediction method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011265363.XA CN112381287A (en) | 2020-11-13 | 2020-11-13 | Chinese developmental reading disorder prediction system and prediction method thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112381287A true CN112381287A (en) | 2021-02-19 |
Family
ID=74583701
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011265363.XA Pending CN112381287A (en) | 2020-11-13 | 2020-11-13 | Chinese developmental reading disorder prediction system and prediction method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112381287A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116936038A (en) * | 2023-09-15 | 2023-10-24 | 北京智精灵科技有限公司 | Visual search training improvement method and system for reading disorder |
-
2020
- 2020-11-13 CN CN202011265363.XA patent/CN112381287A/en active Pending
Non-Patent Citations (1)
| Title |
|---|
| 江显群: "《农业痕量灌溉关键技术研究》", 31 August 2020, 海洋出版社, pages: 110 - 113 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116936038A (en) * | 2023-09-15 | 2023-10-24 | 北京智精灵科技有限公司 | Visual search training improvement method and system for reading disorder |
| CN116936038B (en) * | 2023-09-15 | 2023-12-22 | 北京智精灵科技有限公司 | Visual search training improvement method and system for reading disorder |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110060762B (en) | Psychological development level assessment method and system based on multi-factor scale data | |
| Hecht et al. | The relations between phonological processing abilities and emerging individual differences in mathematical computation skills: A longitudinal study from second to fifth grades | |
| CN109523194A (en) | Chinese reading capability comparison method, apparatus and readable storage medium storing program for executing | |
| CN107657559A (en) | A kind of Chinese reading capability comparison method and system | |
| Barbero-Rubio et al. | Perspective-taking measured by implicit relational assessment procedure (IRAP) | |
| Li et al. | Individual differences in second language acquisition: Theory, research, and practice | |
| KR20110035710A (en) | System and method for studying foreign language based on cognitive funciton of learners | |
| Pavelko et al. | Revisiting reliability: Using Sampling Utterances and Grammatical Analysis Revised (SUGAR) to compare 25-and 50-utterance language samples | |
| CN112381287A (en) | Chinese developmental reading disorder prediction system and prediction method thereof | |
| CN110110066A (en) | A kind of interaction data processing method, device and computer readable storage medium | |
| Atienza et al. | A Deep Neural Network in a Web-based Career Track Recommender System for Lower Secondary Education | |
| KR20130062414A (en) | Evaluation system of learning pattern | |
| Cox et al. | Exploring proficiency‐based vs. performance‐based items with elicited imitation assessment | |
| CN106960400A (en) | Enroll the appraisal procedure of probability, recommend the method and device of college entrance examination universities and colleges | |
| CN114897077A (en) | English reading literacy evaluation method based on machine learning | |
| KR102496412B1 (en) | Operating method for auditory skills training system | |
| CN113919983A (en) | Test question portrait method, device, electronic equipment and storage medium | |
| Hirose | Current Failure Prediction for Final Examination using Past Trends of Weekly Online Testing | |
| CN114091964A (en) | Supplier multi-index evaluation method based on error inverse propagation algorithm | |
| Anderson et al. | An evaluation of the technical and methodological aspects of the School Entry Assessment kit | |
| CN112015780B (en) | Intelligent proposition analysis processing method and system based on deep learning | |
| Jingning | Speech recognition based on mobile sensor networks application in English education intelligent assisted learning system | |
| Rungrat et al. | M-learning platform for assessment and personalized learning of thai language by primary school children | |
| Schuff | Human-centered explainable artificial intelligence for natural language processing | |
| CN111915227B (en) | Teaching evaluation data analysis method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 100101 courtyard 16, lincui Road, Chaoyang District, Beijing Applicant after: INSTITUTE OF PSYCHOLOGY, CHINESE ACADEMY OF SCIENCES Address before: 100083 courtyard 16, lincui Road, Daxing District, Beijing Applicant before: INSTITUTE OF PSYCHOLOGY, CHINESE ACADEMY OF SCIENCES |