Description:Field of Invention
Personalized e-learning system helps to understand an individual’s interest, learning ability, and skills to enhance student knowledge and build confidence among them. To solve problems in the generalized 'teaching-learning method, the study is undertaken to develop a personalized e-learning system using Artificial Intelligence. Right from manufacturing, medical, e-commerce to education, and in every sector, AI plays an essential role in the automation of various activities. Artificial intelligent techniques help to transform learning by providing a personalized approach to the students. Artificial intelligence real power is to store a large amount of data about learners and analyze it and provide personalized education as per individual needs.
Background of the Invention
Learning is a dynamic process that depends on the learner's interest, emotions, existing domain knowledge, and learning ability. The e-learning material is published over the internet in audio, video, presentations, text, discussion forums, webinars, etc. E-learning resources help learners to gain knowledge and skill as per their requirements. The personalized e-learning approach is learner-centric which helps to provide an appropriate learning path for learners. E-learning material is available on the internet, but it is challenging for the learner to select appropriate e-material, satisfying his/her learning expectations. The modern academic world has a fundamental issue in creating a personalised online learning environment. In order to boost student achievement, previous studies have proposed a theoretical framework for customised online learning environments. Intelligent e-learning systems, on the other hand, are built without taking into account their learning behaviour, a crucial component of any e-learning system. Customisation of instruction makes use of methods grounded in expert systems. Instead of creating an interactive system to tailor instruction to each student's unique background and interests, we rely on machine learning algorithms and methodologies to gauge their potential. AI-powered grading and knowledge assessment tools help educators save time while delivering precise outcomes. With the use of artificial intelligence, an adaptive learning system may determine what students need to learn and then tailor its lessons to meet those needs by avoiding unnecessary repetition of concepts that students do not yet understand or by clearing up any confusion between different ideas. With the support of AI, students can overcome learning disabilities and progress through courses at their own speed. Methods based on artificial intelligence can be used to track students' progress. It may change the way teachers work in the future by making it easier for students to work together and integrate what they learn. Some AI-based approaches for handling complicated problems include the fuzzy system, neural networks, genetic algorithms, and deep learning (WO2024080796A1).
Artificial Intelligent in e-learning provides real-time querying and can be work as an assistant for providing knowledge resources. AI plays a vital role in gamification-based e-learning to create learning interest among the learners. AI imitates emotion enough to create a more engaging experience because learning is an emotional process where gaining new knowledge is happiness. According to Beulah Christian Latha a personalized e-learning framework to identify learning patterns, objects, learning styles, and learning paths for the learner was developed. FSML is implemented using association rule mining to identify the learning style of learners. Apriori algorithm is implemented to identify learners' knowledge level concerning learning object complexity level and learners' knowledge. The genetic algorithm technique is used for learning path optimization. Content-based filtering, collaborative filtering, and hybrid approach are used for learning path recommendation. One more focus is on a learner-centric approach to e learning. It uses synchronous and asynchronous collaborative learning techniques. To achieve these learners centric approach, correlation between tutors, learning behavior, learning content and questionnaires is a critical factor in effective learning. It implements the calibration process to match the expected and the actual outcome. It helps to define new learning rules and strategies in the e-learning environment. To forecast the initial skill competency skill of learner gray theory and multi-object programming are used. The fuzzy membership function is used for evaluating the learner’s knowledge level. Data mining algorithms and appropriate AI techniques are necessary for a personalized collaborative e-learning environment. Artificial Intelligent in e-learning provides real-time querying and can be work as an assistant for providing knowledge resources. AI plays a vital role in gamification based e-learning to create learning interest among the learners. AI imitates emotion enough to create a more engaging experience because learning is an emotional process where gaining new knowledge is happiness.
The adaptive teaching pedagogy is one of the best strategies that exploit the interaction between learner and teacher to provide education as per the individual learners' learning style. The Bayesian network technique is used to provide interaction between domain knowledge concept and learners' learning requirements. The adaptive learning system is an intelligent tutoring system that provides a personalized education environment that dynamically adapts learning goals, provides guidance and knowledge, and decides an individual learning path. It is a learner-centric environment that organizes learning material and manages learning strategies. To achieve this a system named Learning Vista to enhance concepts quality of learning in the large class of students is developed. The objective of this system is to provide learning material as per their knowledge level to avoid misconceptions. This provides sufficient opportunities for students to improve their skills and knowledge. Learning optimization is implemented until the learner gets sufficient knowledge. Kolb's learning theory is used to identify learners learning style (US11775850B2).
V. Kavitha, Resham Lohani elaborates on the role of artificial intelligence to enhance the virtual learning environment in e-learning. They suggested implementation of an instructional design model called ADDIE (Analysis, Design, Development, Implementation, and Evaluation) to design courses, content and evaluate learning experience. System design (UI) plays an essential role in motivating and retaining the interest of the learner. Professionals suggested using the Fleming model to integrate different learning tools like Audio learning, visual learning, read/write learning, and kinesthetic learning for an interactive e-learning system. According to their, artificial intelligence plays an important role to facilitate the diagnostic process that identifies students behavior, level of learning, module suggestion as per prerequisite, assessment of learning in the continuous interval, finding the difficult area in learning, manipulate various multimedia presentations according to the learning style of a learner and adjust teaching pedagogy as per student requirements. AI helps to identify the cognitive behavior of learners to support the personalized learning experience
Most of the educationalist focuses on learning style identification in traditional or classroom teaching. Very few ones focus on the e-learning system. Learning style identification is an important factor in a personalized e-learning system. FSLM, VARK these models motivate to identify the learning behavior of the learner. The learning style is dependent on the static and dynamic factors of the learner. According to the existing investigation in education system learner’s aptitude, academic performance, interest, skill, and learning attitude all these factors are necessary to consider while identifying learner‟s learning behavior. It is beneficial to develop a personalized e-learning system that works according to the learning behavior, knowledge level and learning capacity of the learner. The performance, accuracy of the e-learning system can be improved by providing an optimal learning path and recommending suitable e-learning material for the learner. Learner performance evaluation is needed for developing an adaptive e learning system.

Summary of the Invention
A personalized e-learning system is developed, that consist of three modules. This model focuses on learners behavior analysis, learning style identification and domain knowledge prediction of e-learning material along with learning path. To identify the learning behavior of the individual learner a learner’s dataset is created to store academic information, reading, listening ability, and aptitude related information of the individual learner. A fuzzy rule-based system is developed to categorize learners according to their learning behavior into two types as fast and slow learners. To identify the learning style of the individual learner questionnaire is used to collect learner‟s information related to four different learning styles as Auditory, Reading, Visual and Kinesthetic. Artificial Neural Network technique is implemented for predicting the bimodal learning style of learner. To predict existing domain knowledge of learner, the learners existing knowledge data is collected using three types of tests as the basic level, intermediate level, and advanced level tests. Fuzzy rule-based system is used for data analysis. Adaptive Neuro-Fuzzy System (ANFIS) approach is used to predict learner’s domain knowledge level as Unknown, Partially Known or Completely Known.
Brief Description of Drawings
Figure 1: Personalize E-learning system Architecture
Figure 2: Proposed Fuzzified Rule Based System
Figure 3: ANFIS for Domain Knowledge Prediction

Detailed Description of the Invention
Learning behavior analysis is a fundamental challenge in today's education system. It depends on many factors like learner's interest, skill, hobbies, perception, attitude, aptitude, and emotions. It is a dynamic factor in the learning environment. Analysis of learning behavior helps for selecting teaching pedagogy as per learner's need and interest. To develop a personalized e-learning system, it is crucial to identify learners' learning behavior to provide the best learning experience. Fuzzy logic-based systems can be used to handle noisy, distorted and imprecise data. It gives a novel intuitive approach for solving complex problems. Fuzzy logic is based on human communication and based on a quantitative technique used in ordinary language, which is straightforward to use and understand. Learning behavior varies from learner to learner. It is a dynamic factor in learning and challenging to predict the exact behavior of the learner. To solve this imprecise problem fuzzified rule-based system is developed to identify the learning behavior of the leaner. Figure 1 shows a proposed model of a Fuzzified Rule-Based System for learner behavior analysis.
Learner's data is collected using four tabs. Here the student first selects the personal information tab and entered academic information. In the second tab, the students select pdf files for reading a specific topic and then attempt the respective test. Students select the video file in the third tab to listen to a video lecture on a particular topic and attempt the respective test. In the fourth tab, students attempt aptitude tests. Learners Data is extracted from “.csv” file. This file is a collection of learner’s information like PRNNO, Personal Information, and Time taken by the learner to attempt reading, listening, and aptitude test and the score of each test, respectively. Ms Excel macros are used for data cleaning, preprocessing and normalization. Each test performance is measured from test score, and the learner takes total time for attempting the aptitude tests. Fuzzy inference system is a widespread technique based on the fuzzy set theory. Working memory stores all facts related to the input. IF-THEN rules are designed to drive conclusions from facts. Basic fuzzy design consists of components as rule base, comprising fuzzy rules, a database that defines fuzzy rules, and a reasoning mechanism that executes inference procedure. FIS is designed from learner’s academic information collected as a part of personal information, reading, listening, and aptitude test score. The Matlab fuzzy toolbox is used to develop fuzzified system for learner behavior analysis. Since the trade-off between simplicity and concept capture ability is quite good, the triangular membership function is used. The Rule base and database are the important components of the knowledge base. The database provides essential data for fuzzification, rule-base, and defuzzification modules. While designing FIS for decision support system (DSS), domain experts input is considered and K-means algorithm applied on dataset. FIS is built by forming 70 IF-THEN rules. It includes the membership functions that represent the implications of linguistic values of input, output variables, domain-ranges, labels, and shapes.
There are different methods of learning style identification. Identify Learning styles of learners help improve learning interest and make it easy to recommend learning material to the individual learner for selecting appropriate learning methodology for better learning outcomes. Students access teaching resources as per their learning preference leading to an increase in their level of knowledge, metacognition, self-confidence, and motivation. In personalized education, every student learns differently. The method they absorb, process, comprehend, and retain knowledge is different from learner to learner. Example: While teaching program coding, some students jump in and learn by doing and enjoying debugging, testing processes. Some students learn by listing instructions through watching video lectures, some of them learn from observing their friend's work, whereas others may use a hybrid approach for learning too. E-learning focus on four learning styles, namely- Auditory, Reading, Visual and Kinesthetic, etc. Sixteen questions are designed with four questions for each learning style under the guidance of education domain expert. A five-point Likert scale technique is used for identifying learners preference for specific questions. Choices are weighted as Strongly Disagree (1), Disagree (2), Not Sure (3), Agree (4), and Strongly Agree (5), respectively. The questionnaire was uploaded on the Moodle web portal. Individual learners attempted all four questionnaires by login in to the Moodle web portal. The output of the questionnaire was downloaded and stored in a result file. It consists of each of four learning style identification test score out of 20, and the same are exported to a.csv file.
The auditory learner learns better through listening to lectures, discussions, etc. The music helps these learners to improve concentration and memorizing information. These learners like to participate in classroom activities and ready to ask questions for better understanding. The questionnaire is designed to identify the score of the auditory style of each learner. Reading Learners learn better by reading information from various sources. Learner prefers text material for gaining and memorizing information. Reading helps learners to release stress. They like to read books, notes, e-material, text messages, information broachers, newspapers, etc. These learners prefer to participate in reading activities and learn new things by reading books or manuals. The questionnaire is designed to identify the score of the reading style of each learner. The visual learner learns better through imagination and prefers to use pictures, graphs, drawings, illustrations, presentations, etc. They understand and retain the information by visualizing things. They can solve complex problems through visualization. They memorize information by writing it. They prefer to solve puzzles and participate in maze interactively. The questionnaire is designed to identify the score of the visual style of each learner. The kinesthetic learner learns better by doing things practically. They generally perform body movement while speaking or memorizing information. They prefer demonstrations, case studies, and assignments while studying. They learn from group activities, projects, role plays, and tactile process. The questionnaire is designed to identify the score of the kinesthetic style of each learner.
The activation function is a mathematical equation used to generate an output of the neural network. It also helps to normalize the output of each neuron that ranges between (0, 1) or (-1, 1). Logistic Sigmoid is a nonlinear Activation function for binary classification problems. Output raised by function ranges between (0, 1). The package “nnet” is implemented to trained a neural network using an Logistic Sigmoid Activation function and generalized weights. Neural network topology is presented with generic plot() function by using an “nn” class object. Prediction of new observations is done by applying predict() function on dataset. Variables passed to the predict() function are similar in order to those in the neural training network. Artificial Neural Network with a single-hidden-layer neural network is used for a training network. The network is trained by changing the number of hidden neurons. The network is checked for accuracy and for each iteration error is computed. The proposed neural network has four input neurons that indicate learning styles: Auditory, Reading, Visual and Kinesthetic, four hidden nodes as H1, H2, H3, H4, and six output nodes indicate bimodal learning style as AK, AR, RK, VA, VK, and VR. Two bias input nodes B1 and B2 are added into the hidden and output layers respectively. Weight Calculation Formula: (No. of Input Neurons * No. of Hidden Neurons) + No. of Hidden Neurons+ (No. of Hidden neurons * No. of Output Neurons) + No. of Output Neurons. Here “nnet” function uses an algorithm called BFGS (Broyden–Fletcher–Goldfarb–Shanno) optimization to find the internal weight constants. The BFGS algorithm belongs to quasi-Newton methods of optimization.
Domain knowledge identification is a stepwise process as shown in figure 2. Here the system is developed to predict existing domain knowledge of learner by implementing the following stepwise approach: 1. Domain expert design three tests as basic, intermediate and advanced to identify learner‟s knowledge level. 2. Learner login to the system and attempt each test. The score of all three tests generated through the system. 3. Learners score for every three tests are combined into one single .csv file. 4. Domain experts' knowledge base is used to develop Fuzzy rule-based system (FIS) 5. Result generated through the FIS is used as the training dataset. 6. ANFIS (Adaptive Neuro-Fuzzy System) is trained using the training dataset 7. FIS generated through ANFIS is applied on the testing datasets to predict learner‟s domain knowledge as Unknown, Partially Known and Completely Known. Fuzzy logic has been developed using fuzzy set theory. It is basically developed from infinite level logic. There is a correspondence between classical logic and fuzzy logic. It allows us to represent vague concepts expressed in natural language. The membership function assigns the value to elements in a specified range. The value indicates the grade of the membership function of the element. The greater membership value shows a larger degree of membership. The most frequently used membership function having values ranges between the unit interval [0,1]. The Fuzzy sets are completely and uniquely defined by one particular membership function. It not only depends on the concept but the context. There are two types of fuzziness one is intrinsic and another is informational fuzziness. Fuzzy logic deals with linguistic variables and linguistic modifiers. Prepositional logic is also used while developing inference rules. Fuzzy decision making is an effective technique for solving complex problems. The problem in which we can‟t exactly predict the output as true or false (1/0) can be solved using fuzzy logic, as it deals with uncertainty efficiently. Human knowledge can be easily interpreted using fuzzy logic. It is one of the critical AI approaches for solving problems using an explicit representation of human experience. Fuzzy logic can be further used to create a rule base using if-then rules with appropriate function mapping. Triangular membership function is used for developing a model for predicting domain knowledge of learner. Triangular membership function can be defined as y=trimf(x, parameters) where y is output membership value, x is an input variable and parameters means values for membership function.
Rule base is designed by domain expert to categorized learner according to his/her domain knowledge level into three categories like Unknown (UK), Partially Known (PK) and Completely Known. There were total 27 rules. Here some rules are defined among 27 rules of Domain Knowledge FIS as follows: If learners score for the basic level test is less (i.e. score ranges between 0 to 4), the intermediate level test is less (i.e. score ranges between 0 to 4) and the advanced level test is medium (i.e. score ranges between 3 to 7) then learner’s domain knowledge is predicted as Unknown (UK). If learners score for the basic level test is Medium (i.e. score ranges between 3 to 7), the intermediate level test is High (i.e. score ranges between 6 to 10) and the advanced level test is Medium (i.e. score ranges between 3 to 7) then learner’s domain knowledge is predicted as Partially Known (PK). If learners score for the basic level test is high (i.e score ranges between 6 to 10), the intermediate level test is high (score ranges between 6 to 10) and advanced level test is Medium (i.e score ranges between 3 to 7) then learner‟s domain knowledge is predicted as Completely Known (CK). Accordingly remaining rules are also defined to categorized learner as per their 125 domain knowledge level into UK, PK and CK.
The Adaptive Neuro Fuzzy system is implemented to enhance the accuracy of domain knowledge prediction. Takagi and Sugeno coined the Adaptive Neuro-Fuzzy System (ANFIS) model, which obtained advantages of both ANN and fuzzy logic both in one framework. The ANFIS is an automatic learning model through a training dataset. However, this model cannot elaborate details about how it obtains the rules for decision making. On the other hand, the fuzzy logic can produce output out of a fuzzy logic decision, but it cannot automate learning. An ANFIS, derived from nature, automatically creates input and output data pairs from the training dataset. It has been effectively used in various fields related to nonlinear problems. The structure of ANFIS training is, as shown in Figure 3. The ANFIS model is working using a feed-forward neural network with a multi-layer structure. In general, the layers of the ANFIS model is created from six layers. Layer 1: First layer (Input layer) accept the input from two input parameters x and y. Respective input data is passed to the number of neurons in the next layer . Layer 2: Every node in Layer 2 as a circle node and labeled as π. These nodes multiply input signals and send data to output node. Layer 3: Each node in the layer three labeled N is a circle node. The ith node calculates the ratio of ith rule is a summation of all rules which are fired. Layer 4: In the layer four each node i is indicate as square node with node function. Layer 5: In this layer single circle node labeled ∑ that computes overall output as the summation of all input nodes. , Claims:The scope of the invention is defined by the following claims:

Claim:
The Design of A Personalized E- learning system by using Artificial Intelligent Techniques comprising the steps of:
a) identification of learning behavior performance of a learner in terms of reading, listening, aptitude, remembering capacity, hobbies, interest, study methodology.
b) identify the learning behavior of the individual learner related to four different learning styles as Auditory, Reading, Visual and Kinesthetic.
c) predict existing domain knowledge of learner for developing an interactive personalized learning system.
2. As per claim1, A fuzzy rule-based system is developed for learning behavior analysis of learners using the Mamdani approach, K-means clustering algorithm and experts‟ knowledge base system.
3. As per claim1, a logistic sigmoid activation function with one hidden layer is designed for identification of the learning style of the individual learner.
4. As per claim1, an Adaptive Neuro-Fuzzy System (ANFIS) approach is designed to predict learners domain knowledge level as Unknown, Partially Known or Completely Known.