Detecting Anomalies in Students‟ Results Using Decision Trees

Salami, Hamza Onoruoiza; Ibrahim, Ruqayyah S.; Yahaya, Mohammed Onimisi

doi:10.5815/ijmecs.2016.07.04

Cited by 7 publications

(8 citation statements)

References 15 publications

(17 reference statements)

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…In their experiments, they tested four different techniques for anomaly detection in physical spaces covered by wireless networks. Salami et al [20] presented that decision tree can be used to detect anomalies in student performance. Cui and Wang [21] proposed a hybrid model that combines polynomial regression and Gaussian distribution to detect anomalies in electricity consumption data from different schools.…”

Section: Related Workmentioning

confidence: 99%

Multi-Agent-Based Unsupervised Detection of Energy Consumption Anomalies on Smart Campus

Zhang

Xia

2019

IEEE Access

View full text Add to dashboard Cite

The smart campus is becoming a reality with the advancement of information and communication technologies. For energy efficiency, it is essential to detect abnormal energy consumption in a smart campus, which is important for a ''smart'' campus. However, the obtained data are usually continuously generated by ubiquitous sensing devices, and the abnormal patterns hidden in the data are usually unknown, which makes detecting anomalies in such a context more challenging. Moreover, evaluating the quality of anomaly detection algorithms is difficult without labeled datasets. If the data are annotated well, classical criteria such as the receiver operating characteristic or precision recall curves can be used to compare the performance of different anomaly detection algorithms. In a smart campus environment, it is difficult to acquire labeled data to train a model due to the limited capabilities of the sensing devices. Therefore, distributed intelligence is preferred. In this paper, we present a multi-agent-based unsupervised anomaly detection method. We tackle these challenges in two stages with this method. First, we label the data using ensemble models. Second, we propose a method based on deep learning techniques to detect anomalies in an unsupervised fashion. The result of the first stage is used to evaluate the performance of the proposed method. We validate the proposed method with several datasets, and the experimental results demonstrate the effectiveness of our method. INDEX TERMS Anomaly detection and evaluation, energy consumption, multi-agent, smart campus.

show abstract

Section: Related Workmentioning

confidence: 99%

Multi-Agent-Based Unsupervised Detection of Energy Consumption Anomalies on Smart Campus

Zhang

Xia

2019

IEEE Access

View full text Add to dashboard Cite

show abstract

“…This current literature review provides an investigation to the research published in the field of ICT [14,15] and educational menagment [2,7,12], identifies some key issues and emerging messages, and suggest priority areas for the research study undertaken [14].…”

Section: Literature Reviewmentioning

confidence: 99%

“…Key emerging messages [1][2][3][4][5][6][7][8] include the following: the importance of ICT to society and to future prospects is clear within the educational literature, but the real significance of this for educational management has yet to be seen within this research study. The educational management literature is beginning to indicate that attitudes towards ICT are shifting, and often significantly, which is likely to have a major impact upon ICT and school management.…”

Section: Literature Reviewmentioning

confidence: 99%

See 1 more Smart Citation

Analyses of Impacting Factors of ICT in Education Management: Case Study

Fetaji¹,

Fetaji²,

Ebibi³

et al. 2018

IJMECS

View full text Add to dashboard Cite

Research studies and attitudes towards ICT use in education management are shifting, and often significantly. This is likely to have a major impact upon ICT and education management. However, the real significance of the impact for educational management has yet to be seen within this research study. The focus of the research study is to investigate and analyses the ICT usage in Education Management. ICT in teaching has an important role and its impact on the advancement of educational processes related to effective teaching and learning, and modern research in this field is almost irreplaceable. Another important element is the use of different software platforms that facilitate learning visible and make it more concrete, more practical and applicable to everyday life. In order to analyze the data a combination of qualitative and quantitative methodology has been used. In order to analyze this, Case study analysis of high schools in city of Skopje, Macedonia is realized. Insights and recommendations are provided. Index Terms-ICT impact in education, Management of education, evaluation of technology enhanced educationAnalyses of Impacting Factors of ICT in Education Management: Case Study

show abstract

“…Tal constatação é corroborada por pesquisa de Drakakis et al (2016) que identi ca a de ciência do J48 em lidar com dados contínuos, podendo apresentar uma redução em até 20 pontos na acurácia do modelo. Entretanto, para bases que contavam apenas com variáveis discretas Salami et al (2016) identi cou uma acurácia sempre superior a 90%.…”

Section: A Caracterização Dos Dados Afeta a Acurácia Da Classi Cação/unclassified

Avaliação da performance do algoritmo J48 para construção de modelos baseados em árvores de decisão

Vieira¹,

Neves

Oliveira

et al. 2018

RBCA

View full text Add to dashboard Cite

ResumoAs árvores de decisão são modelos hierárquicos utilizados em várias áreas do conhecimento por sua capacidade preditiva e de resolução de problemas de maneira simples e objetiva. Entretanto, apresentam algumas limitações relacionadas à sua adequação à base de dados e ao se atentar quanto aos procedimentos para seleção dos parâmetros de crescimento e poda a serem adotados. Desta forma, têm-se como objetivo avaliar e discutir a performance do algoritmo J48 para construção de modelos de tomada de decisão em árvore em base de dados com atributos de diferentes tipos. Para tanto, realizaram-se experimentos em 10 bases de dados disponíveis em repositório internacional, considerando como variantes os métodos de treinamento, teste e poda, aplicados em toda base de dados e com o uso dos métodos Wrapper e CFS (Correlation-based Feature Selection) para seleção de atributos. Identi cou-se que na presença de dados contínuos, os únicos modelos que apresentaram boa capacidade preditiva estiveram presentes em situações em que a grande quantidade de exemplos puderam compensar tal de ciência. Os modos de treinamento "validação cruzada" e "divisão por porcentagem" mostraram-se similares em suas predições quando ajustados a 10 folds e 75%, respectivamente. Ademais, a seleção de atributos não foi capaz de gerar melhores predições denotando que tal método de forma isolada não compensa possíveis inadequações nas bases de dados. Pode-se constatar que os resultados referentes à capacidade preditiva dos modelos são fortemente direcionados pelo quantitativo de exemplos pertencentes à base, presença de dados contínuos e de dados com ruído.Palavras-Chave: Árvore de decisão. J48. Modelo de decisão. AbstractDecision trees are hierarchical models used in several areas of knowledge due to their predictive capacity and problem solving in a simple and objective way. However, they present some limitations related to their adequacy to the database and in regard to paying attention to the procedures for selection of growth and pruning parameters to be adopted. In this way, the objective is to evaluate and discuss the performance of the J48 algorithm for the construction of tree decision-making models in databases with attributes of di erent types. Therefore, experiments in 10 databases available in international repository were carried out, considering as variants the training, testing and pruning methods, applied throughout the database and using the Wrapper and Correlation-based Feature Selection (CFS) methods for attribute selection. It was identi ed that in the presence of continuous data, the only models that presented good predictive capacity were present in situations in which the large number of examples could compensate for such de ciency. The cross-validation and percentage split training modes were similar in their predictions when adjusted to 10 folds and 75%, respectively. Furthermore, the selection of attributes was unable to generate better predictions denoting that such a method, in an isolated way, does not compensate for possibl...

show abstract

Detecting Anomalies in Students‟ Results Using Decision Trees

Cited by 7 publications

References 15 publications

Multi-Agent-Based Unsupervised Detection of Energy Consumption Anomalies on Smart Campus

Multi-Agent-Based Unsupervised Detection of Energy Consumption Anomalies on Smart Campus

Analyses of Impacting Factors of ICT in Education Management: Case Study

Avaliação da performance do algoritmo J48 para construção de modelos baseados em árvores de decisão

Contact Info

Product

Resources

About