Predicting students’ academic performance by mining the educational data through machine learning-based classification model

Abstract

Students’ academic performance prediction is one of the most important applications of Educational Data Mining (EDM) that helps to improve the quality of the education process. The attainment of student outcomes in an Outcome-based Education (OBE) system adds invaluable rewards to facilitate corrective measures to the learning processes. Furthermore, the explosive increase of e-learning platforms generates a large volume of data that demands the extraction of useful information using up-to-date techniques. Keeping this view in mind and to check the impact of various features on student outcomes during online classes, we have analyzed two sets of datasets; the Kalboard 360 dataset (a larger dataset) that contains academic, demographic as well as behavioral features which have been observed and recorded during the classes held and a local Institute dataset that does not acquire behavioral features. To achieve this, we have selected a few machine learning algorithms such as Decision Tree (J48), Naïve Bayes (NB), Random Forest (RF), and Multilayer Perceptron (MLP) to classify the students, along with a few filter-based feature selection methods like Info gain, gain ratio, and correlation features have been applied to select the key attributes. Finally, we have fine-tuned the learning parameters of MLP called “Opt-MLP” to get an optimized output and compared it with other classification models. Our experimental results conclude that Opt-MLP proves its superiority over other classification models by predicting an accuracy of 87.14% without the feature selection (WOFS) and 90.74% accuracy with the feature selection (WFS) method for data set 1 and an accuracy of 79.37% without feature selection and 97.08% with feature selection for dataset 2. But, when the students’ behavioral feature is considered along with other features, the RF model provides 100% accuracy justifying that students’ behavior during class hours has a great impact on attaining the students’ outcomes.

Appendix

• Confusion Matrix: It presents the complete performance of the model and presents output in a matrix form by calculating the actual values and predicted values as shown in Table 1.

  Example of a Confusion Matrix

  

  • True positive is defined as the predicted value equal to the actual value and both are positive.

  • True negative is defined as the predicted value equal to the actual value, but both are negative.

  • False-positive presents the difference between the predicted value and actual value. The model can predict a positive value but the actual value is negative.

  • False-negative implies that the predicted value is negative, but the actual output is positive.

• Accuracy: It is a ratio between numbers of correctly predicted instances to the total number of instances. It is given by.

$$\text{Accuracy}=\frac{\text{Sum of diagonal of Confusion matrix}}{\text{Sum of the confusion matrix}}$$

• Precision: It is a ratio between correctly classified positive values to total predicted positive values.

$$\text{P}\text{r}\text{e}\text{c}\text{i}\text{s}\text{i}\text{o}\text{n}\left({\text{C}}_{\text{k}}\right)= \frac{\text{T}\text{P}}{\text{T}\text{P}+\text{F}\text{P}}$$

• Recall: Recall is a measure of the ratio of correctly predicted positive classes to all classes in the actual class.

$${\text{Recall}(\text{C}}_\text{k})=\frac{\text{TP}}{\text{TP}+\text{FN}}$$

• Root Mean square error (RMSE): It is expressed as the square root of the mean of all the errors i.e. the square difference between actual values and predicted values.

$$\text{RMSE}=\sqrt{\frac{\left({\text{actual}}_\text{i}-{\text{predicted}}_\text{i}\right)^2}{\text{Total predictions}}}$$

• ROC area: Receiver Operating Characteristics (ROC) is one of the most used parameters for evaluating an ML model. It is a graph between the true positive rate (TPR) and the false-positive rate (FPR). The area under this curve is called the ROC area.