Student-Performance-Prediction-using-Data-Mining-Techniques
Introduction
Here, I have tried to identify and evaluate the impact of the Covid-19 pandemic and its subsequent fallout in predicting student’s academic performance.
For this, a data set of various undergraduate students was compiled from March 2021. A Likert-type questionnaire was administered and large number of responses were gathered from various primary and secondary resources.
This was subsequently used to validate the proposed methodology. Furthermore, different classification algorithms were applied and compared with one another based on the accuracy of their final results.
The results show that the excessive use of e-learning tools including smartphones, laptops and tablets have a significant impact on student performance as well as on their psychological health.
Data Mining Techniques used
-
Decision Tree - Due to its lesser complexity and simple architecture, decision trees are widely used in EDM. It consists of nodes and leaves where data is continuously split in accordance with certain parameters. Another advantage of decision tree is that it requires lesser time for data preparation and is easy to interpret.
-
Random Forests - They on the other hand perform classification by constructing a number of decision trees at training time using bootstrapping, random subsets of features and average voting. It is more robust than decision trees and lesser prone to overfitting.
-
K-nearest neighbors - It determines the class of the data point through a majority voting principle. It means that a class label can assigned to a data point based on it’s distance to it’s nearest neighbors. Its relatively lesser computation time compared to other classification techniques and clarity makes it very useful in EDM.
-
Support Vector Machine- It is generally used for smaller datasets and hence perform relatively faster. In this technique a hyper plane is drawn which helps to separate two or more different classes. The decision boundary or Hyperplane is estimated by maximizing the distance between different groups. The dimension of the hyperplane depends upon the number of features of the class. Due to it’s shorter computation time it’s often used to predict student performance and in other EDM techniques.
-
ANN - It is a biologically inspired programming method that is a popular EDM technique. It is a collections of neurons where each node is connected with one another via links. Each link has a certain weight associated to it which determines the influence of one node over another.
Link to blog : https://towardsdatascience.com/evaluation-and-prediction-of-students-academic-performance-during-covid-19-40bb2b90141b