Desktop version

Home arrow Computer Science

  • Increase font
  • Decrease font


<<   CONTENTS   >>

INTELLIGENT LEARNING ANALYTICS USING DATA MINING TECHNIQUES

With the advancement in the educational field, applications are being developed, where teachers, parents, and students are digitally connected in a forum. The feedback is obtained by teachers from parents and students on classroom activities and teaching aids. These inputs act as intelligent learning analytics for machines. The application helps the parents and students hi the following ways: (a) classroom activities performed are available to parents on mobiles, (b) tracking of students through mobiles, (c) classroom updates are sent online, (d) parents are updated about children activity through reminders, (e) access to teachers resources to improve student outcomes, and (f) manage all the activities online on your accounts and get msights on how other teachers interact with their classrooms.

4.6.1 STUDENTS PERFORMANCE PREDICTION USING DECISION TREES

A decision tree is a mining technique, where predictions are performed after analyzing the datasets. It uses a tree-like structure for formulating predictions. The synthesized educational dataset is shown in Table 4.1. The attributes are stated as follows: (a) class performance, (b) sessional marks, (c) attendance, (d) assignment, (e) lab work, and (f) class grade. The class result of students is predicted after analyzing the attributes using the decision tree technique. The classifier model is shown in Table 4.2, which finds that the "Sessional Attribute" is having the highest Information Gain.

TABLE 4.1 Educational Data Set Description

ID

Class

Performance

Sessional

Marks

Attendance

Assignment

Lab Work

Class Grade

1

Good

Average

Poor

Good

Poor

C

2

Average

Good

Good

Good

Good

A

3

Good

Good

Average

Average

Average

В

4

Poor

Good

Average

Poor

Average

c

5

Good

Poor

Good

Average

Average

в

n

11

n

n

n

n

n

TABLE 4.2 Classifier Information

J48 pruned a tree for “Sessional” Attribute, i.e., Highest Information Gain

SESSIONAL = GOOD

| CLASS PERFORMANCE = GOOD: A(45.0)

| CLASS PERFORMANCE = AVERAGE: A(35.0)

| CLASS PERFORMANCE = POOR: C (20.0)

SESSIONAL = AVERAGE

| CLASS PERFORMANCE = GOOD: A (48.0/8.0)

| CLASS PERFORMANCE = AVERAGE: A (16.0/8.0)

| CLASS PERFORMANCE = POOR: В (16.0/8.0)

SESSIONAL = POOR

| CLASS PERFORMANCE = GOOD: C (20.0)

| CLASS PERFORMANCE = AVERAGE: C (40.0)

| CLASS PERFORMANCE = POOR: C (40.0)

4.6.2 MINING EDUCATIONAL DATA USING KMEANS CLUSTERING

A'-means clustering is the unsupervised technique for classifying the data into clusters. In AT-means, the data are not having any specific class. The A'-means clustering algorithm finds the к number of centroids and designates every input data point to the nearest cluster.

In machine learning, the АГ-means approach helps in (a) pattern matching, (b) finding clusters which are having similar characteristics, (c) advantageous for clustering of voluminous datasets, and (d) efficient way to scale down the size of data. The AT-means clustering approach has been implemented on the dataset shown in Table 4.1, and the results are generated in the form of clusters shown in Figure 4.1. The clusters show (a) the students who are short of attendance and (b) students who have performed poorly in sessional. In the same way, the АГ-means approach can help identify the students as per their learning and understanding skills hi the classroom teaching.

Visualization of attributes clusters

FIGURE 4.1 Visualization of attributes clusters.

4.6.3 EDUCATIONAL DATA MINING USING NEURAL NETWORKS

Neural network self-organizing maps (SOMs) clusters the student’s data into classes based on their class performance, considering the same educational dataset as shown hi Table 4.1. The network after getting trained shows the plots for self-organizing map neighbor distances, weight planes, and sample hits. The results displayed in Figure 4.2(a) and (b) show the cluster of students belonging to a particular class. The classes representing similar features represent the area of neurons with a large number of hits as compared to another region. The SOM distance is shown by calculating the Euclidian distance of each neurons class from its neighbors.

(a) SOM sample hits, (b) SOM neighbor weight distances

FIGURE 4.2 (a) SOM sample hits, (b) SOM neighbor weight distances.

The neural network classifies the input shown in Table 4.1 into a set of target class for pattern recognition, hi neural networks, training is done multiple times to get the refined results and sampling. The desired network class is shown in Table 4.3. There are training, validation, and testing of the samples shown in the dataset.

TABLE 4.3 Desired Target Dataset

1

1

1

1

0

0

0

0

0

0

0

-

N

0

0

0

0

1

1

1

1

1

1

0

-

N

0

0

0

0

0

0

0

0

0

0

1

N

In the training phase, the neural network is trained, and the network is adapted conforming to its error. In the validation phase, network generalization is done, and the training of the network is stopped when generalization stops improving and is concluded.

The testing phase is an absolute part of network performance during and after training. The training performance of the network is shown in Figure

4.3 (a), and the confusion matrix is shown in Figure 4.3(b). The results show that the network has learned to classify data properly with few misclassifications.

(a) Training performance of network, (b) Confusion matrix

FIGURE 4.3 (a) Training performance of network, (b) Confusion matrix.

4.6.4 ACADEMIC COUNSELING OF STUDENTS USING ASSOCIATION RULE MINING (ARM)

With the help of ARM, the students’ preference for the courses to undergo industrial training can be predicted. These techniques become important in academic counseling of students. The semisynthesized dataset collected fr om the students of engineering background is shown in Table 4.4.

TABLE 4.4 Dataset of Optional Courses

Student Ш

Courses Opted(X)

Preferred Course(Y)

1

Core Java, Advance Java

Android Computing

2

Core Java, Advance Java, HTML

Android Computing

3

HTML, JavaScript

PHP

4

C, C++

DotNet

5

Advance Java, Core Java

Android Computing

6

C, C++

Core Java

7

C, C++, C#.Net

Core Java

8

C, C++

Core Java

9

Web Programming

PHP

10

HTML, Web Programming

PHP

n

n

11

In ARM, support and confidence are the two values obtained. The strong rules are extracted from the dataset if they satisfy minimum support and minimum confidence values. The best rules extracted using Apriori and Predictive Apriori Algorithm are shown in Tables 4.5 and 4.6.

4.6.5 PREDICTING TEACHING PEDAGOGIES USING K NEAREST NEIGHBOR TECHNIQUE

The ЛГ-nearest neighbor technique is the supervised data mining technique. In this learning technique, the classification of data is done from the majority of the nearest neighbor of each point. The distance measures used for calculation are (a) Euclidean, (b) Minkowski, and (c) Chebychev distance measures. The sample dataset shown in Table 4.7 consists of a group of students categorized on the parameters as follows: (a) programming skills, (b) aptitude, (c) technical and logical skills, and (d) practical skills.

Courses Opted(X)=Core Java, Advance Java 1 => Preferred Course(Y)=Android Computing 1 conf:(l)

Courses Opted(X)=Core Java, Advance Java, HTML 1=> Preferred Course(Y)=Android Computing 1 conf:(l)

CoursesOpted(X)=HTML, JavaScript 1 ==> Preferred Course(Y)=PHP 1 conf:(l)

Preferred Course(Y) = DotNet 1 => Courses Opted(X)=C,C++ 1 conf:(l)

CoursesOpted(X)=Advance Java, Core Java 1 ==> Preferred Course(Y)=Android Computing 1 conf:(l)

CoursesOpted (X)=C,C++.DotNet 1 ==> Preferred Course(Y)=Core Java 1 conf:(l)

CoursesOpted(X)=Web Programming 1 ==> Preferred Course(Y)=PHP 1 conf:(l)

CoursesOpted(X)=HTML, Web Programming l ==> Preferred Courses(Y)=PHP 1 conf:(l)

Preferred Course(Y)=C++ 1 => Courses Opted(X)=C 1 conf:(l)

Comses Opted(X)=C 1 ==> Preferred Courses(Y)=C++ 1 conf:(l)

TABLE 4.6 Best Rules Discovered by Predictive Apriori

Courses Opted(X)=C.C++ 3 ==> Preferred Course(Y)=Core Java 2 acc:(0.6787)

Preferred Course(Y)=Core Java 3 => Courses Opted(X)=C,C++ 2 acc:(0.6787)

Courses Opted(X)=Core Java, Advance Java 3 ==> Preferred Course(Y)=Android Computing 3 acc:(0.6787)

The results are predicted for the new group of students in Table 4.9 after calculating Euclidean distance, as shown in Table 4.8. The nearest neighbors using the Euclidean distance metric are shown in Figure 4.4.

TABLE 4.7 Dataset Information

Group ID

Programming

SkiUs(lO)

Aptitude(lO)

Technical and Logical Skills(lO)

Practical

Skills(lO)

Grade

G1

8.5

9

9

8.5

A

G2

8

8.5

7.5

7

A

G3

9.5

4.5

8

8.5

A

G4

6.5

5

6

6

В

G5

6

6.5

7

5.5

В

G6

4.5

2.5

3

4.5

С

Gn

3.5

5.5

4.5

5

с

Group

ID

Programming

Skills

Aptitude

Technical and Logical Skills

Practical

Skills

Grade

Euclidean

Distance

Position

G1

8.5

9

9

8.5

A

2.783

Second

G2

8

8.5

7.5

7

A

3.5

Third

G3

9.5

4.5

8

8.5

A

2.345

First

G4

6.5

5

6

6

В

5.196

Fourth

G5

6

6.5

7

5.5

В

5.220

Fifth

G6

4.5

2.5

3

4.5

С

9.565

Seventh

Gn

3.5

5.5

4.5

5

С

8.215

Sixth

TABLE 4.9 Target Class Featured for the New Group

Group ID

Programming

Skills

Aptitude

Technical and Logical Skills

Practical

Skills

Grade

G 7

9

6.5

8.5

9.5

A

4.6.6 PREDICTIONS USING SVMS AND BAYESIAN CLASSIFICATION

The techniques such as SYM and Naive Bayesian play an important role in finding new educational traits. The SYM is a supervised data mining technique, which finds the nearest data vectors, called support vectors nearest to the hyperplane constructed by SVM, whereas Bayesian classification is also supervised learning, which finds the probability of events occurred. Both of these techniques help in improving educational trends and effective decision-making. Using both these techniques, students’ placement results can be easily predicted.

EDUCATIONAL DATA MINING USING PYTHON LANGUAGE

In educational field, machine learning plays an important role in achieving the following: (a) helps faculty to improve their lecture plans by determining where cluster of students are struggling, (b) predict student’s conduct, (c) helps in developing systems which can provide regular feedback to teachers, students, and parents about how the students leant, the learning pace of student, the support required to achieve learning outcome, (d) grade assessments of student’s performance, (e) customize the learning schedule for each student in the classroom, (f) additional assessment to students, (g) smarter learning aid for each student, (h) organize content effectively for identifying weak students, (i) improve retention by identifying “at risk” students, (j) prepare schedules for students and teachers as per then needs, time, and availability, (k) adaptive and interactive tests that help students master each chapter, (1) personalized and experimental learning, (m) programs tailored to every student learning speed and need, (n) visualization of topics help students retain difficult concepts and terminologies, (o) interactive and engaging learning modules; the visually rich content enables conceptual clarity and lifelong term retention, and (p) strategies for seamlessly integrating technology into class.

4.7.1 MACHINE LEARNING FIELDS

Python programming language is one of the prominent fields of machine learning. It includes libraries such as PyTorch for popular deep learning frameworks. There are six classes in PyTorch that can be used for natural language processing (NLP)-related tasks using recurr ent layers.

With NLP, machines are now capable of recognizing and understanding language just like humans. Examples are (a) chatting via text and (b) semantic search. Artificial intelligence and machine learning can transform education. Machine learning helps (a) teachers to focus on eveiy student during the course teaching, (b) education for the specially abled students is possible through machine learning, and (c) grading assessment of the students. Machine learning helps us find patterns in data, and from that, predictions about new data points are made. To get those predictions right, we must construct the data correctly. There are some of the commonly used libraries in Python shown in Table 4.10 for machine learning.

TABLE 4.10 Commonly Used Libraries for Data Science

Library

Utility

Pandas

Library for data analysis

Statsmodels

It is the library for statistical modeling

scikit-leara

It is the library for classification and clustering using data mining techniques

NumPy

It is the library for Numerical routines

SciPy

These libraries are used for scientific and technical computing

matplotlib

It is the plotting library for NumPy

NLTK

It is the commonly used library for Statistical Natural Language Processing

Keras

It is Neural Network and Deep Learning Library

TensorFlow

It is the primary tool for deep learning analytics and an open-source AI Library

4.7.2 DATA DESCRIPTION

The predefined libraries of Python Machine Learning language are implemented on the dataset described in Table 4.11.

In the dataset, numerical grading is done on the internal assessment marks given by faculty members for three different subjects. The dataset is synthesized for experimentation purpose. The subjects for which numerical grading is done are (a) PC Packages, (b) Microsoft Access, and (c) VB.NET.

The input data obtained using predefined data mining libraries in Python have been shown in Table 4.12.

The classification of students’ data is done on the following: (a) Marks obtained in Sessional, (b) Attendance, and (c) Assignment Submission.

Criteria

Internal Assessment

Remarks

Reg_id

651429

Random Enrollment No. of Student

Internal

Marks

PC Packages (20)

Microsoft Access (20)

YB.NET

(20)

Marks obtained by students on the basis of Internal Assessment Criteria defined.

Obtained

17

13

14

Subjects: {PC Packages. Microsoft Access, YB.NET}

Sessional Marks (10)

8

4

6

Student marks in Sessional test

Attendance Marks (5)

4

4

4

Punctuality in class

Assignment Marks (5)

5

5

4

Assignment submitted by students

N Students

N

N

N

TABLE 4.12 Data Representation Using Python Libraries

Sessional Marks (10)

Attendance Marks (5)

Assignment Marks

(5)

Grading, i.e., Target Class (20)

Predictor

variables

Predictor

variables

Predictor

Yariables

Dependent variable i.e. Taiget class

PC Packages

MS Access

Л-B.NET

PC Packages

MS Access

Л-B.NET

PC Packages

MS Access

T3.NET

Target Class {Grading: “A” —Good"“B”

—Average “C” —Poor}

8

6

6

5

4

4

5

3

3

AC C

7

8

6

3

3

4

5

5

5

В В В

6

6

3

4

5

3

3

3

3

СВ С

7

8

9

4

4

4

3

2

5

В В А

5

9

8

2

5

5

3

4

4

С А А

N

N

N

N

N

N

N

N

N

NN N

4.7.3 RESULTS AND DISCUSSIONS

Python generates the following plots for three different subjects: (a) univariate plots, (b) multivariate plots, and (c) histogram plots. The input dataset in Table 4.11 is processed using python machine learning. These plots help understand the attributes of relationship and distribution. The results in the form of plots are displayed in Figures 4.5-4.7. The results have shown the following: (a) student’s maximum rate of interest among three subjects, (b) understanding level of the subject, (c) assignment performance, and (d) attendance ratio of students among three subjects.

CONCLUSION AND FUTURE SCOPE

In this chapter, machine learning and its emerging roles in educational data mining are discussed. In this sequence, the predefined libraries of Python machine learning language commonly used in data science in the present scenario for effective decision-making and intelligent learning analytics are discussed. With these predefined libraries, implementation of data mining techniques and deep learning concept of neural network in Python can be used with more accuracy and effectiveness.

Furthermore, in this direction, intelligent learning analytics over educational domain has been implemented to find meaningful information in the form of knowledge from voluminous data. The classification and clustering techniques discussed in this chapter are as follows: (a) decision trees, (c) naive Bayesian, (d) neural networks, (e) nearest neighbor, (f) AT-means, and (g) ARM. Using these techniques, the predicted results obtained are as follows: (a) cluster students having shortfall of attendance, (b) identifying students who have performed poorly in sessional, (c) prediction of probability of students placements, (d) guiding students to choose appropriate courses to undergo industrial training, (e) discovering new pedagogical practices of teaching, (f) improving student skills and increasing employment chances for them, (g) strengthening decision-making of institution to include industry-oriented courses into curricula, (h) guiding the students to explore other skills such as aptitude, reasoning, and communication apart from regular studies to increase placement opportunities, and (i) strengthening of in-house training activities of faculty members.

Another focus of this chapter is that using Python, multiple classifiers have been implemented on an educational dataset to obtain the results. With the implementation of predefined Python libraries for data mining, the results

have been obtained, for example, the maximum interests of students for the course among other courses of the curriculum have been predicted.

The results obtained using different data mining techniques and machine learning in the proposed chapter help in efficient analysis and prediction of new academic trends for the betterment of students. In future, research can be used as follows: (a) admission seekers to analyze and choose the most suitable courses, (b) generate skill-oriented manpower, (c) fulfilling the skill gap of industry and academia, (d) develop educational software and visualization techniques for prediction of student’s performance in examinations, (e) cluster those students who need exclusive focus in course, and (f) best utilization of unutilized and unstructured data to get the meaningful results.

KEYWORDS

  • cognitive
  • education
  • information
  • intelligence
  • supervised
  • unsupervised

REFERENCES

  • 1. B. Bakhshinategh, O. R. Zaiane. S. Elatia, and D. Ipperciel. “Educational data mining applications and tasks: A survey of the last 10 years,” Educ. Inf. Technol, vol. 23, no. 1, pp. 537-553,2018.
  • 2. P. Dillenbourg, “Collaborative learning: Cognitive and computational approaches,” in Advances in Learning and Instruction Series. Amsterdam, The Netherlands: Elsevier Science, 1999.
  • 3. S. Slater, S. Joksimovic, V Kovanovic, R. S. Baker, andD. Gasevic, “Tools for educational data mining: A review,” J. Educ. Behav. Statist., vol. 42, no. 1. pp. 85-106, 2017.
  • 4. Y. Zhang, S. Oussena, T. Clark, H. Kim. “Use data mining to improve student retention in higher education—A case study,” in Proc. 12th Int. Conf. Enterprise Inf. Syst., 2010, vol. l,pp. 190-197.
  • 5. I. Lykourentzou, I. Giaimoukos. V. Nikolopoulos, G. Mpardis, and Y. Loumos, “Dropout prediction in e-leamiug courses through the combination of machine learning techniques,” Coinput. Educ., vol. 53, no. 3, 950-965, 2009.
  • 6. S. B. Kotsiantis, C. J. Pierrakeas. and P. E. Piiitelas, “Preventing student dropout in distance learning using machine learning techniques,” hi Proc. hit. Conf. Knowl.-Based Intell. Inf. Eng. Syst., 2003, pp. 267-274.
  • 7. X. Zhu, “Machine teaching: An inverse problem to machine learning and an approach towards optimal education.” in Proc. 29th AAAI Conf. Artif Intel!., 2015.
  • 8. J. Liebowitz, “Thoughts on recent trends and future research perspectives in big data and analytics in higher education,” in Big Data and Learning Analytics in Higher Education. Cham, Switzerland: Springer, 2017, pp. 7-17.
  • 9. S. B. Alier and L. M. R. J. Lobo, “Combination of machine learning algorithms for recommendation of courses hi E-Learning System based on historical data,” Knowl.- Based Syst., vol. 51, pp. 1-14, 2013.
  • 10. J. E. Beck and В. P. Woolf. “High-level student modeling with machine learning,” hi Proc. Int. Conf. Intell. Tutoring Syst. 2000, pp. 584-593.
  • 11. C. Dede, “The evolution of distance education: Emerging technologies and distributed learning,” Amer. J. Distance Educ., vol. 10, no. 2, pp. 4-36,1996.
  • 12. J. Demsar, B. Zupan. G. Leban. and T. Curk, “Orange: From experimental machine learning to interactive data mining,” in Proc. Ear. Conf. Principles Data Mining Knowl. Discov., 2004, pp. 537-539.
  • 13. T. M. Mitchell, “Machine learning and data mining,” Commun. ACM, vol. 42, no. 11, pp. 30-36, 1999.
  • 14. R. S. Baker and P. S. Inventado, “Educational data mining and learning analytics,” hi Learning Analytics. New York, NY, USA: Springer. 2014, pp. 61-75.
  • 15. C. Bmgos, M. L. Campanario, D. de la Pena, J. A. Lara, D. Lizcano, M. A. Martinez, “Data mining for modeling students’ performance: A tutoring action plan to prevent academic dropout,” Coniput. Elect. Eng., vol. 66, 541-556, 2018.
  • 16. J. Naren, “Application of data mining hi educational database for predicting behavioural patterns of the students,” Int. J. Coniput. Sci. Inf. Technol.,x ol. 5, no. 3, pp. 4649-4652, 2014.
  • 17. M. Abdous, H. Wu, and C.-J. Yen. “Using data mining for predicting relationships between online question theme and final grade,” J. Educ. Technol. Soc., vol. 15, no. 3, pp. 77-88, 2012.
  • 18. R. S. Baker, “Educational data mining: An advance for intelligent systems in education,” IEEE Intell. S)!St., vol. 29, no. 3, pp. 78-82, 2014.

APPENDIX

Apriori: It is an association rule mining algorithm to find the frequent item sets.

ARM: ARM means association rule mining. It is the data mining technique that shows the probability of relationships in the form of if-then statements among data items in a large dataset.

Bayesian classification: It is based on Bayes theorem and represents a supervised learning approach for classification of data. This technique follows the probabilistic approach.

Confusion matrix: In machine learning, a confusion matrix displays the classifier performance in a tabular format.

Decision tree: It is a data mining technique and displays the tree-like structure of decisions. This technique also follows the probabilistic approach. J48: J48 is an algorithm for decision tree analysis and performs predictions. Л'-means: It is a clustering technique of data mining. АГ-means follows unsupervised learning approach, and it is used to find groups in the data.

Neural network: It is a computer system based on nervous system just like in human brains. It is the learning algorithm and can be supervised or unsupervised.

NLP: NLP means natural language processing. It is a field of artificial intelligence, where a large amount of natural language data is processed and analyzed.

Supervised learning: It is the learning approach where learning is performed using given training data, and there is response, that is, decision variable. The classification is an example of it.

Support vector machines: It is a supervised learning algorithm of data mining. In SVMs, data analysis is performed for classifications and regression analysis. In SVM, there is a separating hyperplane, which separates the hyperplane into two parts to lay class on either side of the hyperplane.

Unsupervised learning: It is the learning approach where learning is performed without past training data and no decision variable. Clustering is an example of unsupervised learning. In this learning approach, the data given are not labeled, that is, only input variables are given with no output variables.

 
<<   CONTENTS   >>

Related topics