Random Forest Based Classifiers for Detecting Result Anomalies

Stanley Ziweritin,  Iduma Aka Ibiam

Random forest(RF) is a supervised machine learning approach that experts use to build and integrate many decision trees into a single forest. It takes considerable expertise to detect result anomalies depending on the degree of disparity between students CA and exam scores. It is doable to train RF-based classifiers to accurately identify anomalies with imbalanced data categorization. The aim is to develop RF-based classifiers capable of detecting abnormalities in student results, such as when a student performed remarkably well on the exam but poorly on the CA, or vice versa. The SMOTE technique was used to resolve unbalanced data categorization, which helped reduce dataset bias toward the majority class while also ensuring that the minority class received an acceptable sample size. Strong decision-makers were grouped into a class of majority vote using the grid search and randomized function. Trees capacity to learn from small data samples was arbitrarily constrained by the uniformly distribution function, which increased model accuracy and reduced tree correlation. Comparatively, the Classification, Adaboost, and GradientBoosting classifiers produced accuracy scores of 99.00%, 95.17%, and 81.50% respectively

Title:	Random Forest Based Classifiers for Detecting Result Anomalies
Author Name:	Stanley Ziweritin, Iduma Aka Ibiam
Abstract:	Random forest(RF) is a supervised machine learning approach that experts use to build and integrate many decision trees into a single forest. It takes considerable expertise to detect result anomalies depending on the degree of disparity between students CA and exam scores. It is doable to train RF-based classifiers to accurately identify anomalies with imbalanced data categorization. The aim is to develop RF-based classifiers capable of detecting abnormalities in student results, such as when a student performed remarkably well on the exam but poorly on the CA, or vice versa. The SMOTE technique was used to resolve unbalanced data categorization, which helped reduce dataset bias toward the majority class while also ensuring that the minority class received an acceptable sample size. Strong decision-makers were grouped into a class of majority vote using the grid search and randomized function. Trees capacity to learn from small data samples was arbitrarily constrained by the uniformly distribution function, which increased model accuracy and reduced tree correlation. Comparatively, the Classification, Adaboost, and GradientBoosting classifiers produced accuracy scores of 99.00%, 95.17%, and 81.50% respectively
Cite this article:	Stanley Ziweritin, Iduma Aka Ibiam , " Random Forest Based Classifiers for Detecting Result Anomalies " , International Journal of Emerging Trends & Technology in Computer Science (IJETTCS), Volume 11, Issue 6, November - December 2022 , pp. 009-018 , ISSN 2278-6856.
Full Text [PDF] Home