Subject: Submission of report on "Enhancing Undergraduate Business Student Performance through Education Data Mining: A Study on Classification, Regression, and Pattern Analysis.". I am writing to formally submit my report on "Enhancing Undergraduate Business Student Performance through Education Data Mining: A Study on Classification, Regression, and Pattern Analysis," which was assigned as part of my BBA program in United International University. This is a research paper that focuses on educational data mining and its impact on improving the education system in Asian countries.
The article highlights the importance of using data mining to improve the quality of education, improve student performance and bring about positive changes in the education system. The scope of the study is to use a general framework to predict the CGPA of business students using a DNN approach. The objectives of the article are to predict student performance, find efficient patterns and related data to search for the most important courses for the final result.
The authors intend to use classification methods such as Random Forest, Gradient Boosted Tree, Tree Ensemble, Decision Tree, SVM and KNN to predict the performance level of students. Overall, the paper highlights the importance of education data mining and its impact on improving the education system in Asian countries.
Introduction
This version of student categories creates great difficulties for their faculties to discover their students who are truly at risk for their academic standing. To solve these problems, we use a generalized framework that otherwise helps us to predict the CGPA of students using DNN. We will use a student data set which was collected student transcripts of business background students from a recognized university in Bangladesh.
Prediction performance alone cannot achieve the objective of success for improving the educational performance of students or the quality of the teaching procedure. Students and faculty can easily understand their problems through this early research of predicting educational performance. Most of the students are frustrated in their study and go on the opposite path of study which helps this type of student exactly.
This algorithm helps to understand the faculties for predicting the educational performance of their students and what they should do for their students to overcome their bad path to the right path. They predict the quality of their students and what steps their students need to take from this educational data mining process algorithm.
Literature Review
Data mining analysis is typically based on three techniques: (1) inferential statistics (2) artificial intelligence and (3) machine learning. Data mining benefits from these technologies, but differs in the objective pursued: extracting patterns, describing trends, and predicting behavior. The study of the foundations of this data mining can be seen as a scientific research in its environment is the scope of the theory of data mining.
Enough attention not to be devoted to data mining to study the environment or its philosophical roots. It is obvious that abstract studies of data mining as a field of scientific research, an alternative collection of samples of isolated algorithms, are useful for the further development of the whole field. Many types of main questions focus on data mining above these layers and together they form a complete epithet of the field.
The layered framework is indisputable by using 3 subfields of data mining, classification, measurements and justified oriented data mining [6]. Introduction with the development and purpose of data mining, many researchers were concerned in the basic issues, the basis of data mining. The relationship between mining data and base of mining data is similar to the relationship between computer science and computers.
For the study of data mining research in the data mining environment and in the field of data mining theory. De-archiving renews a timely opportunity to evaluate algorithms on a large number of data packets. In recent years, most research on this topic has emphasized the use of data mining theory, and some sources have applied some cases of machine search theory, guess the ability of the newest, select the current catalog for the next idea, and discover some of them. recent and relevant studies.
Evaluation data are applied with caution since data mining algorithms must predict success in a program (fail or pass) and learning theory activity must be evaluated based on their predictive quality, ease of study, and user-friendly characteristics. for users. After the data collection was completed, data mining theory was applied after compiling the data. This methodology is used to help students and papers, the method of the international workshop on intelligent teaching systems, the Educational Data Workshop (EDMW).
Research Methodology
Before we can deal with the missing data, we must first learn about the nature of the data that is missing. The tendency for a data point to be missing for use in MAR has nothing to do with missing data, but it has something to do with some of the observed data. There are two likely causes of missing value, both of which depend on the hypothetical value (e.g. low-achieving students in a class do not want to share their CGPA) or missing value depending on the value of another variable (e.g. .Assume that the majority of female students do not want to know how old they are!.The missing value in the age field is influenced by the following factors that vary by gender).
We changed all the grades with their associated grade points to build the second version because our collected data includes grades. We used four unique terminologies to indicate the children's condition, as shown in Table 3.4: 'Excellent', 'Very good'. The students' gender and condition, as mentioned before, are categorical factors included in our data set in this study.
The data is scaled to a predefined extent to bring each of the features to a similar size level. Random forest usually avoids this by randomly selecting subsets of the parameters and using those selections to create smaller trees. A large number of trees can lead to an algorithm that is too slow and useless for real-time predictions. This is the main limitation of the random forest.
The number of correct predictions divided by the total number of input samples is the ratio. As a result, while examining the results for each of the different classes, we have chosen F measures, sensitivity, specificity and precision. The Python programming language was used for all implementations of these three steps.
We used 32 neurons in the second hidden layer because before output it is the last layer. Rectified Linear Unit (ReLU): When creating a DNN model, ReLU [28] is the most commonly used activation function. This measure is used to calculate the squared difference of the predicted and target values for each data point.
One of the important advantages of this metric is that this metric is straightforward and easy to understand because the error is expressed as a percentage. For all the test samples, we used the performance of a decision tree as a baseline.
Experimental Results
As seen in Figure 4-1, Random Forest outperformed the other models with the highest accuracy (94.01%). After completing the training session, we performed applying our proposed method DNN to make some predictions on our dataset. The converging line indicates that the two-axis line has been quite consistent in converging between losses and epochs, indicating a good fit of the model.
After the training and validation phases, run the model which predicts the final CGPA using the test data. We displayed all 45 test cases in the line graph in Figure 4-3 to better understand the variances in actual and predicted CGPA. 4-3 shows that, with the exception of a few places, we see that there is a significant difference between the predicted and actual values.
For the vast majority of samples, the model fits perfectly and accurately predicts the actual value. Based on the pattern test and error measurements, we can infer that the model used has exceptionally well learned the underlying pattern from educational data and can reasonably estimate the real CGPA of undergraduate students. We can see from the tree diagram and rule set that 7 of the 10 students achieved an 'A+' in course '201(c),' and 5 students also achieved 'A+' in course '302.' Our data set contains 335 students who received first grade as their final grade.
Good” and “At Risk” as described earlier to indicate a student's performance level based on each semester's GPA. The number of students achieving the other three states, however, steadily declined, with the "good" trend line declining faster than the other two. We also created a trend chart, which you can see in Figure 4-5, to compare it to the predicted result.
The cross-state analysis after each semester is shown in Figure 4-5, along with the GPA earned for that semester and the expected score. As a result of the aforementioned variables, we can conclude that most Bangladeshi business students perform poorly in the initial period. In the experimental stage, the classification and regression algorithms are evaluated to get a good result by predicting the performance of the students to analyze the actual value and the predicted value.
Conclusion
Finally, in the second stage, we proposed using a DNN model with a lower error rate than traditional models to predict the final CGPA of undergraduate business students. Three evaluation measures, MSE, MAE, and MAPE, were used to evaluate the performance of the DNN model. In addition, compared to the basic decision tree model, our proposed approach reduced the MSE, MAE, and MAPE errors by 6.066, respectively.