Predicting Student Success Using Big Data and Machine Learning Algorithms

Abstract

The prediction of student performance, allows teachers to track student results to react and make decisions that affect their learning and performance, given the importance of monitoring students to fight against academic failure. We realized a system of the prediction of academic success and failure of the students, which is the overall result and the goal of the educational system. We used the personal information of the students, the academic evaluation, the activities of the students in VLE, ​​Psychological, the student environment, and we added practical work and homework, mini projects, and the number of student absences which gives a vision of the quality of the student. Then we applied the methods of artificial intelligence and educational Data mining such as KNN, C4.5 and SVM for the prediction of the academic success of students, but these methods are not sufficient given the progressive number of students, specialties, learning methods and the diversity of data sources as well as student data processing time. To solve this problem, Big Data technology was used to distribute the processing in order to minimize the execution time without losing the efficiency of the algorithms used. In this system we cleaned the data and then applied the property selection algorithms to find the useful properties in order to improve the algorithm prediction rate and also to reduce the execution time. Finally, we stored the data in HDFS and we applied the classification algorithms for the prediction of student success using MAPREDUCE. We compared the results before and after the use of big data and we found that the results after the use of Big Data are very good at execution time and we arrived at a recognition rate of 87.32% by the SVM algorithm.

Analysis of Student Performance Applying Data Mining Techniques in a Virtual Learning Environment

“…The results of this study agree with some previous works, where it is pointed out that intermediate grades [21], forums and participation in classes [15][16] [18], links, and tasks [11] are key factors in academic performance. Regarding the discrepancies, it was obtained that the monthly income had a low relationship with academic performance, unlike the work done in [16].…”

“…In [20], they found variables that make it possible to predict school performance, pointing out that homework, access to the VAS, questionnaires, and age are the main predictor variables. In [21], they looked at the SVM, C4.5, and KNN algorithms for predicting academic success. The results showed relevant factors: final grades, economic status, parental level of education, the distance between home and institution, student interest, and access to the VLE.…”

“…This predictive algorithm is part of the family of trees, which orders the importance of the characteristics of the prediction model, placing attributes: academic, socioeconomic, and interaction with the VLE as independent variables and the attribute Passed the course (yes or no) as the dependent variable. b) Due to the good results reported in previous studies, two prediction models were built using the well-known Random Forest [14] and Support Vector Machine [21] classifiers. Random Forest (RF) comprises a set of decision trees, where each tree predicts a category, and then the most voted category is chosen.…”

Reduction of false negatives in multi-class sentiment analysis

“…SVM can be both linear and nonlinear. Dividing the entire dataset into two categories by clearly drawing a decision boundary or hyper plane is called linear SVM [21]. In those cases, a linear SVM classifier has to be implemented.…”

Electricity Theft Detection using Machine Learning

“…The SVM algorithm is designed to find the optimal separating hyperplane between classes based on support vectors (extremes of the class distributions). The training data are separated into classes using boundaries, which results in the maximization of the distance between the various data sets and the boundary [16].…”