Predict students' dropout and academic success

Identification of at-risk students at the start of their academic journey enables timely interventions.
Providing personalized support to help students reach their academic potential.
Expanding from "fail/success" to include "relative success" for a nuanced understanding.
A model must be built that is not limited to a specific field of study but can be generalized across courses in the institution. Motivation<br>

A dataset created from students enrolled in undergraduate courses of Polytechnic Institute of Portalegre, Portugal. The data refers to records of students enrolled between academic years 2008/09–2018/2019 and from different undergraduate degrees[2].
Each record was classified as Success, Relative Success and Failure, depending on the time that the student took to obtain her degree[2]
Logistic Regression (LR) , Support Vector Machines (SVM) , Decision Trees (DT) and an ensemble method, Random Forests (RF) .
Boosting Classifications :Gradient Boosting, Extreme Gradient Boosting, Logit Boost, CatBoost Past research<br>

Dataset Polytechnic Institute of Portalegre, Portugal
Academic years 2008-09 through 2018-19
4424 instances
36 features
The features consist of:
Demographic data
Application info
Course info
Student info
Results from the first two semesters
Home country info
The response is whether the student dropped out, graduated, or is still enrolled.<br>

Data Categorization<br>

XGBoost is an enhanced version of gradient boosting, which is an ensemble learning technique that pays attention to misclassified results to iteratively improve the performance and minimize loss using the gradient descent function.
We will be using XGBoost Classification technique for our dataset. Although existing literature has used it, our focus will be on improving the accuracy by employing techniques like:
Feature Selection
Data Augmentation to balance the dataset
Hyperparameter tuning Methods<br>

K-means clustering is a popular algorithm for partitioning data into clusters by maximizing similarity inside a group and minimizing similarity outside a group

Existing literature does not try using clustering methods. We will not only try to better understand our dataset through clustering methods (which can also in turn help in feature selection) but also derive inferences about relations between the data. Methods (Cont.)<br>

Intended experiments Feature Selection Some unnecessary features can be omitted, to see if it improves accuracy. Clustering Clustering the features into different groups. Hyperparameter tuning 80-20 split
75-25 split
Different types of methods to resolve overfitting, such as K-fold cross-validation, early stopping, and regularization. Data Augmentation This is to balance the dataset. This might improve the performance.<br>

Metrics Accuracy Correct predictions divided by the total number of predictions across all classes F1 Score Computes The number of times a model made a correct prediction across the entire dataset Recall TP/(TP+FN) : ratio of true positives to the total number of positive samples<br>

Realinho,Valentim, Vieira Martins,Mónica, Machado,Jorge, and Baptista,Luís. (2021). Predict students' dropout and academic success. UCI Machine Learning Repository. https://doi.org/10.24432/C5MC89.
Realinho, Valentim, Jorge Machado, Luís Baptista, and Mónica V. Martins. 2022. "Predicting Student Dropout and Academic Success" Data 7, no. 11: 146. https://doi.org/10.3390/data7110146
https://www.siue.edu/inrs/factbook/
“Fig. 3 a General Architecture of XGBoost.” ResearchGate, www.researchgate.net/figure/A-general-architecture-of-XGBoost_fig3_335483097.
Harezlak, Armando Teixeira-Pinto & Jaroslaw. 2 K-Means Clustering | Machine Learning for Biostatistics. Bookdown.org, bookdown.org/tpinto_home/Unsupervised-learning/k-means-clustering.html. References<br>

Thank you<br>