Creation of an adaptive system of training students based on lms moodle

CREATION OF AN ADAPTIVE SYSTEM OF TRAINING STUDENTS BASED ON LMS MOODLE

Shohida Yusupova Botirboevna
Urgench Branch of Tashkent University of Information Technologies named after Muhammad al-Khwarizmi
Urgench, Uzbekistan
[e-mail]

Sardorbek Davletboyev Zokirjon ugli

Urgench Branch of Tashkent University of Information Technologies named after Muhammad al-Khwarizmi
Urgench, Uzbekistan
[e-mail]
Bahrom Ishmetov Yangibaevich
Urgench Branch of Tashkent University of Information Technologies named after Muhammad al-Khwarizmi
Urgench, Uzbekistan
[e-mail]

Abstract - In the article, methods of obtaining and analyzing results using the KNIME Analytics Platform are considered to build a fuzzy model of assessing student knowledge through adaptive tests, to conduct experiments on the built model. The importance of creating a personalized training course for students based on LMS Moodle is discussed, and the developed method uses a neural network model to classify the current level of user knowledge.
Keywords. Adaptive tests, LMS Moodle, assessment, fuzzy algorithm, expert system, neural network, test bank, preceptor.

1. INTRODUCTION

In accordance with the large flow of information and the currently widespread type of distance education, there is a need for new educational technologies that allow the implementation of individual educational trajectories. The field of information education, in which the preparation of professional knowledge of students is carried out, requires the skills of independent processing of large amounts of data.
Adaptation of the educational process consists in the availability of educational materials of different complexity and the presence of complex control issues of different (simple, medium, high) levels. This is the development of special skills (emotional stability, the nature of introversion, responsibility, speed of information perception, reflection) that arise as a result of the use of distance learning technologies in teaching and distance learning of students without face-to-face contact. makes it possible to solve the difficult task of making connections of the teaching level (high, medium, low).
The traditional system of monitoring and evaluating student learning has advantages as well as a number of disadvantages. The most important of them is the individual behavior of the teacher related to the manifestation of subjectivity in the assessment of knowledge. There are studies that show that the coincidence of the grades of two examiners who independently tested knowledge of the same subjects through an intermediate survey corresponds to no more than 60% cases [1].
The stage of updating the system of monitoring and evaluating the quality of higher education is connected with the use of such a powerful tool as a pedagogical test. The reason for this is that the tests allow to determine the presence of controlled characteristics with sufficient objectivity and reliability, as well as to assess the degree of their formation. Today, test technologies are considered as the main means of quality control of education [2].
Pedagogical test is a tool designed to measure student learning and consists of a system of test tasks, a standardized procedure for processing and analyzing the results. It is no exaggeration to consider it as a system free from the shortcomings of the traditional management system. Currently, there are automated methods of knowledge assessment.
Testing as a kind of control of educational and cognitive activity of students is considered in many didactic works. In modern didactics, the test acts as a convenient and reliable tool for diagnosing and monitoring the educational process. The issues of test technologies are related to the content and methodological content [3].
Adaptive tests are automated, which allow for an individual approach to students, the content of the task, the order of execution, the rule, the score that the student can collect as a result of the completion of this task, and instructions for summarizing the test results.
Adaptive testing makes it possible to provide computer-based issuance of tasks at an optimal, approximately 50% level of the probability of a correct answer for each student. Now in the world there are three variants of adaptive testing. The first is called pyramid testing. In the absence of preliminary assessments, everyone is given a task of medium complexity, and only then, depending on the answer, everyone is given a task easier or more difficult. The second option - flexilevel-control begins with the level of difficulty, which is chosen by the one who is being tested, with a gradual approach to the real level of knowledge. The third option is stradaptive (from the English Stratified adaptive), when testing is carried out using a bank of tasks divided by difficulty levels. If the answer is correct, the next task is taken from the upper level, if the answer is wrong - from the lower one. Thus, an adaptive test is a variant of an automated testing system with previously known complexity parameters and differentiation in the choice of tasks [4].
Adaptive tests can be successfully used in the module-credit paradigm of organizing the educational process. For this, the pedagogue should have the ability to create and practice multiple-choice test tasks of different levels of difficulty on one topic, chapter, section, course content.
It is possible to increase the quality of the test, first of all, its validity, by connecting the given questions with the already received answers. However, even in the simplest case, such a test system is actually an expert system with "if, then..." rules in the knowledge base. Creating such a database is a very difficult task. Adaptive systems of forming and filling the database require organizing the operation of a whole educational complex to study the level of education.

2. 
   METHODOLOGY
   

Based on the results of the test tasks, the system gives a numerical score based on quantitative and percentage indicators. But this assessment does not allow to draw conclusions that give a comprehensive picture of the student's success in this subject. Therefore, it is relevant to develop a plug-in that allows to evaluate the quality of the student's activity and to create a customized training course for the user based on this evaluation. The developed method uses a neural network model to classify the current level of user knowledge. Let's consider how a neural network model is created. The construction of a neural network is carried out in two ways: direct and recurrence distribution. When the neural network is created by direct distribution, then the usual neural network consists of input and output layers. The input layer consists of an input vector and a hidden matrix linking with a hidden layer, and the output layer consists of an output vector and output matrix linking with a hidden layer [5].

The input data for the neural network is the vector of answers after passing the boundary control of knowledge. The output of the neural network provides a fuzzy estimate of the user's level of knowledge. With the help of this assessment and the procedural model described in the article, an optimal set of training tasks is formed. A set of learning elements is selected based on a vague assessment of the level of learning activity of the user. For those elements, the value of the variable "presence" is greater than 0.5, meaning that the element is fully present and placed in the recommended course structure. Using this approach, it is possible to develop a plugin that allows you to create a course structure consisting of a set of elements of educational material with learning activities for a specific student. After creating a new training course structure, the user goes through all the steps mentioned above again. Educational elements necessary for successful development of unlearned materials are selected from the database. Training sessions continue until the user's training level is equal to the level required by the qualitative assessment. Then the course is considered successfully completed.
Let's consider the task of setting a linear test for the assessment of knowledge using artificial neural networks. For example, a linear test has 20 questions, each of which has 5 possible answers. Then the inputs of the neural network are the correct or incorrect answers to all the test questions. Thus, the inputs are binary, i.e. binary: "1" means that the person who passed the test gave a correct answer, "0" means that the answer was chosen incorrectly. The output of the neural network is used to evaluate the student's knowledge.
For example, as usual, knowledge can be evaluated with "unsatisfactory", "satisfactory", "good" and "excellent" grades, then we have four binary outputs of the network. The value "1" corresponds to the rating "unsatisfactory" in the first output, and the value "1" in the last output corresponds to the rating "excellent". The principle of the system is based on machine learning of neural networks. The principle of operation of the system is based on the ability of neural networks to learn. System training consists in changing the weights of connections (weight coefficients of connections between neurons) of network neurons according to given algorithms, called learning rules. The purpose of the training is to compare the grades given by the neural network on the test answers with the grades obtained during the initial test in order to identify the student's level of preparation and his individual abilities.
A multi-layer properly distributed perceptron was used as a neural network, its structure is shown in Fig. 1, in which each neuron transmits its output signal to the input of other neurons, as well as to itself.

Figure 1. Perceptron scheme
Neural network training algorithm. Neural network training is the search for such a set of weight coefficients, in which the input signal after passing through the network provides the result we need. The generalized learning process of a neural network is schematically shown in Figure 2 [6].

Figure 2. Generalized neural network training algorithm.

Perceptron consists of 3 types of elements: S-elements, A-elements and R-elements. S-elements are a layer of senses, i.e. receptors. Each receptor can be in one of two states - resting or excited, and only in the latter state it transmits one signal to the next layer, associative elements - A-elements. They are called associative because each such element, as a rule, corresponds to a whole set (combination) of S-elements. The A-element is activated as soon as the number of signals from the S-elements at its input exceeds a certain value d. The signals from the excited A-elements, in turn, are transmitted to the addition R, and the signal from the i-associative element is transmitted by the coefficient ω_i. This coefficient is called the weight of the A-R connection. Like the A elements, the R element calculates the sum of the input values multiplied by the weights:

there S is the input signal.
Perceptron training consists of changing the weight coefficients of A-R connections.
To experiment with the built model, we will build the following wordflow using the KNIME Analytics Platform (https://www.knime.com/knime-analytics-platform) program (Figure 3).

Figure 3. A model built in KNIME Analytics Platform to experiment with the built model
Reading the first block in this model FileReader (Fig. 4).

Figure 4. Load a selection using FileReader.
Before building the model, we split the dataset into 2, there are training set vs testing set. We separate training set to 80% and 20% to testing set. Training set is given to neural network. (Figure 5).

Figure 5. Configuring MLPLearner before training
As shown in this figure, the number of iterations is 100, the hidden layer is one, and the number of hidden neurons is 20. After training, we pass the model built in ErrorPlot to MultilayerPreseptronPredictor for testing (Figure 6). To make the result more accurate, you can use the CrossValidation testing method.

Figure 6. Test results
A multifaceted perceptron is a perceptron with additional layers of A-elements. The possible number of training samples for a test consisting of 20 questions with 5 answer options is 205 = 32000000. The size of the training sample can be limited to 100 options for answering the test. This is only 0.00003% of the possible answers. Such a sample size is acceptable because there is a clear relationship between the relative error of the test estimation by the neural network and the size of the training sample. This relationship is shown in Figure 7.

Figure 7. A graph of the reduction of errors in the training process of an artificial neural network.
Minimum learning error - 5.0 %, is achieved with a sample of 100 answer options. The 10% accuracy allowed for the test was achieved with approximately 20 training samples. Obviously, the direct training of a neural network for knowledge assessment is not acceptable due to the very large sample number for the accuracy of the test and the minimal error of the answers, that is, the large number of observations. Adaptive learning practices, as well as solutions for combining modeled learning (simulation of different learning situations) with adaptation using fuzzy neural networks are proposed.
To build an individualized training course, it is suggested to use the automated generation system of test tasks based on the results of the diagnostic test and the module for the automated construction of an individualized training course in the LMS Moodle environment.
Customization in LMS Moodle consists of designing a set of learning elements that are optimal for a particular user. The teacher creates a bank of questions for the training course. In the bank, questions are organized into categories. Usually, a separate category is created for each course, and there are also categories that correspond to general course categories. The student registers in the system according to his account. The application is made through a web interface, which allows working with the system from any computer with a browser. He chooses educational courses available to him, studies thematic material. Then the student passes the initial control, which consists of a set of test tasks, and this task is set by the teacher at the initial stage of the course. The uncertainty in the structure of the model is that the set of distinct positions and transitions is related to the fuzzy variable "item presence" and will be a unique set of learning elements for each user.
The considered system consists of separate interacting components. A component is an elementary indivisible block of material that can be: text, web page, file link, web page, task, etc. Each component has its own state. A component's state is an abstraction of relevant information needed to represent its future actions. The state of a component depends on the historical reality of that component and changes over time. It represents the behavior of the system being modeled. Actions of system components are characterized by intermingling or parallelism. The actions of one component of the system can be performed simultaneously with the actions of other components.
III. CONCLUSION
A competent professional will be able to come up with many non-standard ideas, find unusual answers, improve a creative product, find an ideal solution to a problem by adding additional details [7].
Based on the results of access control, the teacher who develops the training course determines the initial structure of the training course, that is, the set of components for the initial presentation of the didactic material to the student. This control shows whether the student's level at the time of the test is low, medium or high. Alternative (substitution) elements that may be present in further changes in the structure of the training course are also identified. Elements defined in the initial course structure are assigned a value of 1. Alternative elements are given a value in the range [0-0,5], which determines only their possible presence in subsequent changes in the study course. A teacher designing a training course is asked to select from a list of possible values for each element the value of the uncertain variable "presence" that determines the allowed coefficient ni: For example, complete: , probably : , slightly: , small: .
The considered approach to building an information adaptive learning system based on LMS Moodle has a number of advantages over competing systems and allows:
- adapting the structure of the training course to the specific user;
- monitoring the delivery of user training sessions based on neural networks;
- conducting additional research in this area in order to improve the quality of computer learning automation.
However, in our opinion, the test cannot replace the exam, preparation for which includes working through the material, memorizing theorems. It develops a culture of speech, the ability to express one's thoughts, develops logical reasoning skills and makes memorization more long-term. A computer test does not allow one to judge these achievements of a student [8].