Diagnosing Learning Disorders in Children: A Comparison of Certainty Factor and Dempster-Shafer Methods

 Abstract —Even though educational technology is very advanced, some children experience learning disorders. Learning disorders in children include Dyslexia, Dysgraphia, Dyscalculia, and Dyspraxia. Ignorance about learning disorders in children will result in the child not getting help to reach his potential and have an impact on problematic behavior and destructive mental disorders in children. That is why it is necessary to make an early diagnosis to determine the presence of learning disorders in children. Therefore, for this reason, this study aims to develop an expert system for the early diagnosis of learning disorders in children using the Certainty Factor and Dempster-Shafer methods. The results show that the Certainty Factor method is more accurate than the Dempster-Shafer method in diagnosing children with disorders. The accuracy of the test results by diagnosing children’s learning disorders using the Certainty Factor method is 90%, and by the Dempster-Shafer method, it is 87%. The novelty of this research is to build a system for diagnosing the types of learning disorders in children using the Certainty Factor and Dempster-Shafer methods which have never been done by previous researchers.


I. INTRODUCTION
Currently, educational technology is developing rapidly [1,2], as well as innovations in the application of educational technology in learning vary widely [3].Besides that educational technology supports today's learning [4].However, some children have learning disorders even though these children have intelligence and sensory average [5].Learning disorders are also not due to intellectual deficiencies, emotional disturbances, or cultural differences [5].Children's learning disorders include reading, writing, and arithmetic [6].The problem for children with learning disorders, such as reading disorders or Dyslexia, is that if reading learning disorders are not treated, the impact of subsequent disorders will also be the same [7].Children who are not recognized or fail to be known to have a learning disorder will struggle to overcome their learning difficulties without the teacher or parents knowing the cause.As a result, it will prevent these children from getting help to reach their potential.Therefore, children who enter school with learning disabilities are at risk of being continuously left behind and further behind in learning compared to their peers.Moreover, according to Colenbrander et al.'s work [8], by knowing early on about learning disorders in children, these children can get appropriate help in acquiring skills and prevent problematic behavior and destructive mental disorders in children.That is why it is necessary to carry out an early detection or diagnosis to find out learning disorders in children, which is very important for early therapy [5,8,9].Keeping in mind that related to problems in learning, learning difficulties or learning disorders become a topic of observation in childhood in the first year of school [10].Still, the main problem is whether learning disorders can be identified early [11].This research makes it happen.That is why for this reason, this study aims to develop an expert system for the early diagnosis of learning disorders in children.
Many factors that cause learning disorders in children include Dyslexia, Dysgraphia, dyscalculia, and dyspraxia [12].Some children show disturbances in mathematics, not due to socio-economic, educational, emotional, psychological, or intellectual factors, which is called disorder dyscalculia [13].In short, dyscalculia is a child's inability to count or learn numeracy or mathematical skills despite average intelligence [14].Impaired ability to spell writing, or spelling errors in the production of text in children, is a dyslexia disorder [15,16].Dyslexia learning disorder represents a visual disability that affects reading ability [17].Moreover, according to Talepasand and Eskandaripour et al.'s research [18], a child's learning disorder due to an inability to recognize words is a dyslexic disorder.Meanwhile, the inability to develop writing skills is typically a dysgraphia failure [16,[19][20][21][22].In contrast, dyspraxia is a disorder associated with motor function in children [23].Meanwhile, according to Pedro and Goldschmidt, dyspraxia is a learning disorder in children who have difficulty carrying out activities [24].
Many prior studies build systems that have the expertise (expert systems) in solving problems in various fields [25], including the system built using the Dempster-Shafer method [25].However, at this time, the Certainty Factor method is starting to be dominantly used as an inference engine because of its ability to construct independent causal assumptions [26] and deal with some rules to produce conclusions [27].
A Dempster-Shafer is helpful in the embodiment of modeling structures [28].Dempster-Shafer or also known as evidence theory [29,30], is a very popular method used in various research [30] and is widely used in multiple applications because the Dempster-Shafer is very flexible and effective in uncertainty modeling [31] and making decisions with uncertainty [32].In essence, the Dempster-Shafer represents independent pieces of evidence [30,32], also known as beliefs [30,33] or the theory of belief function [32,33].This theory provides a framework for embodiment modeling and methods for combining different sets of evidence [33].In short, Dempster-Shafer is a mathematical tool for dealing with uncertainty in attributes [34].Meanwhile, the certainty factor is useful as an inference engine and makes assumptions [6].The Certainty Factor method can realize the level of certainty over identification based on evidence or characteristics [35].In essence, the certainty factor accommodates the uncertainty of an expert in analyzing information [27].This study developed an expert system for early diagnosis of learning disorders in children using the Certainty Factor and Dempster Shafer methods.
The structure of further discussion in this manuscript is as follows.The second subsection discusses previous related work and compares the differences with this work.Meanwhile, the third subsection describes the methodology of this research.The fourth subsection focuses on describing the results and discussing the research results.The last subsection (fifth subsection) discusses the study's conclusions and represents the novelty of the research and suggestions for further investigation.

II. RELATED WORK
Some of the latest related works of prior study are as follows.
Kuerten et al. conducted a literature review on dyslexia learning disorder [36].The previous research was a literature review on dyslexic children's learning disorders.In contrast to the research in this article, this study builds a system for the early detection of learning disorders in children.Thus the difference between the two studies is in the methods and objectives of the study.Pedro and Goldschmidt revealed the teacher's level of understanding of learning disorder dyspraxia and the support needed in teaching children with learning disorders of dyspraxia [24].The difference between this previous research and this article's research lies in the research method and the purpose and object of the research.
Yulianti et al. compared the effectiveness of Dempster Shafer and Certainty Factors in determining adolescent learning styles [37].This previous research has different objectives and objects from this article's research.Likewise, the research conducted by Sari et al. [38] has different objectives and research objects compared to the research in this article.
Waber et al. described learning disorders in children and diagnosed the clinical characteristics of children with dyspraxia disorders in children with dyslexia [23].Previous research is not the same as the research in this article in terms of research methods and objectives.In essence, previous studies diagnosed the characteristics of learning disorder dyspraxia in children with learning dyslexia.In contrast, the research in this article diagnoses children with learning disorders early, either dyspraxia, Dysgraphia, dyscalculia, or dyspraxia.Vlachos and Avramidis perform comparisons to show developmental Dyslexia and developmental Dysgraphia as distinct learning disorders in children [16].The difference between this previous study and the research in this article is that previous studies reviewed dyslexia and dysgraphia learning disorders.In contrast, the research in this article builds a system for early detection of learning disorders in children, both dyslexia and dysgraphia learning disorders, as well as dyscalculia and dyspraxia in children.
Pagliarin et al. [39] assessed children's and adults' special situation anticipatory abilities.This previous research has a different purpose from our study in this article.Previous research has focused on learning disorders or developmental Dyslexia.Instead, the research in this article focuses on the early diagnosis of learning disorders: Dyslexia, Dysgraphia, dyscalculia, and dyspraxia in children.
Safarova et al. [40] conducted quantitative research to assess handwriting ability on learning disorders Dysgraphia.However, this prior study has a different objective, object, and method than this article's study.The previous research was a survey study with the object of learning dysgraphia disorder and aimed to assess the level of dysgraphia learning disorder.In contrast, the research in this article builds a system with objects to detect the type of learning disorder in children using the Certainty Factor and Dempster-Shafer methods.
Meanwhile, Mammarella et al. [41] examined the characteristics of children who have dyscalculia or mathematical learning disorders.Even though this previous study has similarities with the research in this article in examining learning disorders in children, the two studies have different objectives and research methods.Previous research revealed the characteristics of the learning disorder dyscalculia in children.On the other hand, the research in this article diagnoses the presence or absence of learning disorders in children early.
O'Dea et al. synthesized discrete qualitative study findings about the preferences of children and adolescents with developmental coordination disorder with a metaethnographic approach.Previous research compared with the study of this article is different in methods, objectives, and research objects.The previous research method synthesized the literature study results.In contrast, the research method in this article was to build an early prediction system for learning disorders using the certainty factor and Dempster-Shafer methods; likewise, if this previous study synthesized the results of prior research, while the research in this article builds a system to recognize the types of learning disorders in children.
Mustafaeva investigated the phenomenon of learning and teaching disorders in reading and spelling foreign languages [42].Previous research has investigated the phenomenon of learning and teaching disorders in children, in contrast to the research in this article which focuses on developing a system for the early detection of learning disorders in children.Thus, the striking difference between the two studies is in the purpose of the study and the research method.
Taylor and Vestergaard investigated previous studies in psychology and dyslexia neuroscience-related learning disorders [7].This previous study is a literature study of the prior experimental research on dyslexia learning disorders.The previous research has different methods and objectives compared to the research in this article, namely using the certainty factor method and the Dempster-Shafer for early diagnosis of learning disorders in children.
A review of some of the most recent related works (see also Table I The expert system developed in this study for early diagnosis of learning disorders in children shows that the expert system developed using the Certainty Factor method has higher accuracy than the Dempster-Shafer method. Table I shows the differences in previous related studies and their differences from the research conducted in this article.Previous related studies did not test the method/results' accuracy.Not only that, previous research was not a study by building a system like this research and then conducting trials (experiments) on the system built using the Certainty Factor and Dempster-Shafer method in diagnosing learning disorders in children as was done in this research article.

III. RESEARCH METHODOLOGY
The programming language used in building a web-based intelligent system in this study is PHP (Personal Home Page or PHP Hypertext Preprocessor).PHP is the most popular high-level programming language used in building webbased or mobile computer programs [44,45].The stages of research and development of an expert system for early diagnosis of learning disorders in children using the Certainty Factor and Dempster-Shafer methods in this study used or adopted the Waterfall model.The Waterfall is a management model for developing application programs [46].The sequence of processes in the Waterfall model is sequential from the initial stage to the next step [47].Fig. 1 shows a series of development of an expert system for early diagnosis of learning disorders in children in this study.The requirements analysis stage is a step to get the data needed (data collection) to develop an application system.The design stage is realizing the expert system's knowledge base, which contains symptoms of learning disorders and certainty factor scores from experts (see Fig. 2).Meanwhile, the development stage is the stage of applying the research method used or developing the system being built.The testing stage is testing the developed system and whether it follows the needs.If it is inappropriate, then a review of the previous stage is carried out at this testing stage.Finally, the implementation stage is the last stage.At this implementation stage, an expert system was developed to determine the performance of the system being built.

A. Requirement Analysis
Interviews with the expert were obtained to determine the types of learning disorders in children and the accompanying symptoms in this study.The expert interviewed was a pediatrician.The collection of data about the types and symptoms accompanying learning disorders of children was carried out in early 2022 by experts.The symptoms accompanying each type of learning disorder in children, obtained from the expert doctor, are the reference data for the realization of the expert system.The expert system developed works like an expert who can diagnose types of learning disorders in children.Table II shows the types or classifications of learning disorders in children.

B. Design
The design stage is the stage of realizing the knowledge base of an expert system of learning disorders in children with accompanying symptoms and a score of Certainty Factors (CF) obtained from expert doctors (see Table III).Table III contains the knowledge base of the expert system built into this study which consists of symptom codes, the symptoms that accompany learning disorders, and the value of the certainty factor (CF) from experts.

C. Development 1) Certainty factor method
The Certainty Factor method begins by selecting the symptoms and the level of belief in the chosen symptom.The next step is calculating the value of CF (with a single premise).The value of CF is the sum of the user's CF value with the expert's CF value or CF (H, E) = CF (user)  CF (expert).Then, calculate the CF Combine or CF Combine (CF1, CF2) = CF1 + CF2  (1 − CF1).With the completion of all these calculations, the highest value obtained from the calculation of the CF Combine is the final result of this diagnostic process.
The diagnostic process results conclude that the type of learning disorder in children is Dyslexia, Dysgraphia, Dyscalculia, or Dyspraxia (see Fig. 3).An example of a calculation using the Certainty Factor method in diagnosing the type of learning disorder in children with three symptoms that have been given weights by experts and the symptom weights entered by the user is the calculation as shown in Table IV.The next calculation step is to calculate the value of the CF combination obtained from the multiplication in the previous steps.Finally, the results of their combinations are shown in Table V.The result of the diagnosis of the type of learning disorder in children is the type of Dyslexia disorder (P1) with a confidence of 0.928 or 92.8% (See Table V). 2

) Dempster-shafer method
The Dempster Shafer method starts with initialization, then choosing the symptoms that are felt and calculating using the Dempster Shafer formula.The Dempster-Shafer method includes the Belief formula (Bel), which represents a measure of certainty or confidence in the evidence of a set.If the value is 0, it indicates no evidence.On the other hand, a value of 1 means there is a certainty.Plausibility (Pls) is a measure of disbelief or uncertainty against evidence.Value Pls is from 0 to 1.The Belief function and the Plausibility function are: Calculating the value of CF Start Finish

Calculate CF Combine
Enter symptoms and confidence level

Finding the highest value of CF Combine
Summing up the types of learning disorders X = Symptom 1 of the disease.Y = Symptom 2 of the disease.Bel(X) = Belief (X), meaning the belief value or certainty of disease X experiencing symptoms 1. Pls(X) = Plausibility (X), meaning the value of uncertainty or the uncertainty of disease X which experiences symptoms 1. m 1 (Y) = Mass function.Or confidence level of evidence (Y).If it is known that X and Y are a subset, with m1 as a function of density X and m2 as a function of density Y, then m3 is a function of the combination of m1 and m2 (see Fig. 4).The equation of m3 is as follows: An example of the calculation of the Dempster-Shafer method in diagnosing the type of learning disorder in children with three symptoms that experts have weighted is shown in Table VI.
The second step is to calculate the belief and plausibility values in G5, as shown in Table VIII.Next is to calculate the combined function of m1 and m2 using Eq. ( 3) as shown in Table IX.

D. Testing and Implementation
The accuracy of the Certainty Factor (CF) and Dempster-Shafer (DS) methods is tested, which is compared with the results of expert diagnosis using 30 patient (P) data (See Table XII).As presented in Table XII, the Certainty Factor Method correctly diagnosed 27 data from 30 patient data, so the accuracy obtained was 90%.While the Dempster-Shafer method can correctly diagnose 26 data from 30 patient data, the accuracy obtained is 87%.Thus, the accuracy of the Certainty Factor method is better than the Dempster-Shafer method for diagnosing types of learning disorders in children.In short, the results of the test using the Certainty Factor method showed that there were three data on the results of testing which did not match the expert reference data.In contrast, in the results of the test with the Dempster-Shafer, there were four data on testing results that did not correspond with the expert reference data.So, the accuracy difference between the Certainty Factor and Dempster-Shafer testing methods is 3%.In other words, the accuracy of the Certainty Factor method is better than the Dempster-Shafer method in diagnosing learning disorders in children.It happens because the Certainty Factor method, in its calculations, accommodates the expert's weight value and the user's weight value, which then combines the two values to get the result.Here the Dempster-Shafer method only utilizes the value given by the expert, regardless of the user's input value in the calculation [38,49].
The results of this study provide answers to questions from previous researchers who asked whether learning disorders can be identified early [11].The answer is yes, it can, and not just learning disorder dyslexia but other learning disorders: Dysgraphia, dyscalculia, and dyspraxia.Besides that, the results of this study reinforce many previous studies that the development of expert systems helps solve problems in various fields [25], including systems built using the Dempster-Shafer and Certainty Factor methods.The point is that the difference or strength of this research, compared to similar related research, is that this research not only develops a system for diagnosing learning disorders in children but also conducts system trials on data on learning disorders in children.In addition, this study researched all categories of children's learning disorders (Dyslexia, Dysgraphia, Dyscalculia, and Dyspraxia), which previous researchers had never done.Also, this study tested the accuracy of the results of the method used, which previous related studies had not carried out.

V. CONCLUSION
The expert system for diagnosing learning disorders in children using the Certainty Factor and Dempster-Shafer methods found that the system's accuracy using the Certainty Factor method is 90%, and the Dempster-Shafer method is 87%.It means the expert system using the Certainty Factor method is more accurate than the Dempster method.The implication of the results of this study is that it is useful to help identify learning disorders in children from an early age, including learning disorders dyslexia, Dysgraphia, dyscalculia, and dyspraxia, and also at the same time, answer doubts about previous researchers who questioned whether learning disorders in children could be detected from the start.
The novelty of this study is that no previous related research has conducted this research by comparing the Certainty Factor and Dempster-Shafer methods in diagnosing the types of learning disorders in children.Suggestions for further research are to examine various objects other than learning disorders in children and compare them or use different methods.Besides that, it is necessary to carry out further research by paying attention to differences in gender, and age categories (learning level) of children.

Fig. 1 .
Fig.1.The Waterfall model of system development in this study[48].

Fig. 2 .
Fig. 2. Stages of requirements analysis and design of the expert system.

Fig. 3 .
Fig. 3.The process of diagnosing learning disorders in children with the Certainty Factor method.

Fig. 4 .
Fig. 4. The process of diagnosing learning disorders in children with the Dempster-Shafer method.
XI, the result of the diagnosis of the type of learning disorder in children is the type of Dyslexia disorder (P1) with a value of 1 or 100%.
) confirms that the research carried out in this article differs from previous jobs.This study's results help identify children with learning disorders early.So that by knowing early, children with learning Disorders Dyslexia, Dysgraphia, Dyscalculia, and Dyspraxia can get therapy earlier.The novelty of this study is to build a learning disorder detection system in children using the Certainty Factor and Dempster-Shafer methods which previous researchers have never studied.TableIcompares this research work with several prior related researches.

TABLE II :
TYPES OF LEARNING DISORDERS IN CHILDREN

TABLE III :
KNOWLEDGE BASE OF EXPERT SYSTEM OF LEARNING DISORDER

TABLE IV :
THE RESULT OF MULTIPLYING SYMPTOM WEIGHTS FROM EXPERTS AND SYMPTOM WEIGHTS FROM USER INPUT

TABLE VI
The next step is calculating the belief and plausibility values for each symptom in TableVI.But, first, calculate the belief and plausibility values in G1, as shown in TableVII.