Research and application of grade prediction model based on decision tree algorithm

Processo de Mineração de Dados Educacionais aplicado na Previsão do Desempenho de Alunos: Uma comparação entre as Técnicas de Aprendizagem de Máquina e Aprendizagem Profunda

“…Transformação do atributo G3 de numérico para níveis de classificação -nesse procedimento cada faixa de notas recebeu um valor no formato de caractere, atribuindo uma categoria/classe para os registros: notas entre 20 e 16 = "A", notas entre 15 e 11 = "B", notas entre 10 e 4 = "C", notas entre 4 e 0 = "D". Essa abordagem foi inspirada no estudo desenvolvido por Zhang & Wu (2019), os autores tinham como objetivo prever o desempenho de alunos, mais especificamente a previsão de notas dos alunos. Para isso, os autores utilizaram dados de estudantes que cursaram um curso MOOC sobre Programação em Linguagem C, os atributos eram basicamente sobre: informações básicas do perfil dos alunos, pontuação nas atividades avaliativas, número de questões solucionadas, pontuação final, postagens nos fóruns de discussão.…”

“…Corroborando as afirmações quanto à relevância da precisão dos modelos baseados em Árvores de Decisão, Zhang & Wu (2019) afirmaram que tais modelos são consideravelmente simples de serem implementados, e têm precisão relativamente satisfatória, em seu estudo os autores encontraram valores de acurácia variando entre 75% a 81% em modelos baseados em Árvores de Decisão. Por isso, os autores salientaram que estes podem ser empregados para apoiar ações para análise do desempenho de alunos, em diversos cenários educacionais.…”

Prediction of learning outcomes via clickstream data using machine learning

“…These studies were typically carried out at institutional level, where demographic information is readily available, so that appropriate intervention strategies Chapter 5: Review of Grade Prediction Models and Deep Neural Models can be designed and administered to prevent drop-out. To understand the impact of demography on learning outcomes, existing studies involved the use of non-cognitive traits such as their scholastic aptitude [213] and psychosocial aspect of learners [213], personality traits [220], family social economic status [174,207,214], gender and age [174,177,183,198,207,210,[214][215][216][217][218][219], and pre-admission academic achievements [197,207]. Such analysis was performed using correlational analyses and various classification techniques.…”

“…Since learners are expected to perform similarly for courses that they have not taken and those that they had already taken, some studies have also explored the use of historical academic achievements obtained within a program for forecasting [174,177,178,183,188,[196][197][198][199][200][201][202][203][204][205][206][207][208][209][210][211][212]. Furthermore, some studies hypothesized that course structures, concept and topic dependencies between courses and even instructional differences between educators can affect the learning outcomes.…”

“…To quantify the amount of learning effort, interactions within VLEs were considered in terms of the amount of time spent within the platform [172][173][174][175] and frequency of accessing various learning resources such as video content, assessment problems, discussion forums and other collaborative exercises such as microblogs. Video watching behaviors were also analyzed through the frequency of various video playback actions (play, pause, rewind, fast-forward) [176][177][178][179], proportions and the total duration of video played [175][176][177][178][182][183][184], the total number of views [176][177][178][179], and video watching frequency and intervals [180,181]. While video content forms the primary source of information [119], due to the lack of face-to-face opportunities [175,179,184,186,190,191], the topics that they had posted or replied within discussion boards [192][193][194][195], their contribution quantified using social network analysis features such as their centrality, eccentricity, authority and hub scores [173,[187][188][189] and the sentiments of their posts [185,186] were employed to determine if learners would complete the course.…”