Loading...
10 results
Search Results
Now showing 1 - 10 of 10
- PREVISÃO DE ABANDONO DE ALUNOS NUMA INSTITUIÇÃO DE ENSINO SUPERIORPublication . Sobreiro, Pedro; Martinho, DomingosO abandono é um problema nas instituições de ensino superior que tem vindo a aumentar a sua visibilidade, onde se verifica pouca investigação recorrendo a técnicas de Machine Learning. Neste estudo procuramos desenvolver um modelo para prevermos o abandono dos alunos, utilizando os dados históricos dos alunos de uma instituição de ensino superior. Os dados foram utilizados nos algoritmos de classificação Two class logistic regression, Two class boosted decision, Two class neural network e Two class support vector, onde avaliarmos a sua exatidão através da matriz de confusão e análise do Receiver Operating Characteristic curve. Os resultados obtidos permitiram identificar Two class neural network como o mais adequado para os dados que estamos a tratar. No entanto, verificamos que necessitamos de aumentar a representatividade do abandono na amostra e incorporarmos mais variáveis, como satisfação com a instituição e oportunidades de trabalho.
- ANÁLISE DE CLUSTERS PARA SEGMENTAÇÃO DE ESTUDANTES NUMA INSTITUIÇÃO DE ENSINO SUPERIORPublication . Sobreiro, Pedro; Martinho, DomingosA segmentação do mercado é um tema importante para os administradores das instituições de ensino superior. A segmentação dos alunos permite a diferenciação e a definição de ações personalizadas de acordo com cada segmento e pode ser realizada recorrendo a dados existentes de alunos para serem posteriormente utilizados no desenvolvimento de ações de comunicação ou para realização de um acompanhamento interno diferenciado. A metodologia utilizada para realizarmos a segmentação dos alunos (n=280) recorreu à análise de clusters utilizando o algoritmo k-means disponível na biblioteca scikit. O k-means é um algoritmo não supervisionado para a determinação dos clusters, que requer que o investigador determine à priori o número de clusters pretendidos, utilizando uma aproximação iterativa calculando o centro ótimo de cada cluster. A identificação do número de clusters foi baseada no método elbow, que utiliza o pressuposto de que o número de clusters ótimo é aquele em que adicionando mais clusters não reduz significativamente a variância entre clusters. Depois de obtivermos cada cluster realizamos a sua caraterização utilizando as variáveis existentes para termos uma melhor compreensão dos dados. Os resultados obtidos permitiram identificar três clusters, onde obtivemos no cluster um 89 alunos, cluster dois 16 alunos e cluster três 175 alunos. Para facilitar a compreensão dos resultados obtidos realizamos a redução das variáveis existentes através de do Principal Components Analysis, uma redução de dimensões para podemos projetar os dados num espaço dimensional menor de duas dimensões, num gráfico de dispersão x,y. Realizamos a caraterização (média±desvio padrão) das variáveis idade, ano, estado civil e sexo. Os resultados obtidos evidenciam que nos clusters um, dois e três as médias de idades são aproximadamente iguais 28,29 e 31, o estado civil é maioritariamente solteiros com 80%, 81% e 75% e o sexo feminino representa 49%, 51% e 50% respetivamente. Os resultados conseguidos não são elucidativos considerando os indicadores obtidos em cada cluster. Para podermos retirar melhores conclusões deveriam ser incluídas mais variáveis, como cursos frequentados, resultados obtidos na frequência do curso e aumentar a amostra. Um aspeto que poderia ter sido equacionado seria a normalização dos dados, reduzindo impacto de variáveis em escalas diferentes na determinação do número de clusters. Por último seria interessante explorar as diferenças entre os alunos nos clusters existentes realizando a análise das variáveis existentes.
- Sales forecast in an IT company using time seriesPublication . Sobreiro, Pedro; Martinho, Domingos; Pratas, AntonioThe sales forecast is fundamental for the planning of the activity of the companies providing, important indicators for the support of the decisions of the managers. This study aims to explore the potential of time series prediction algorithms in an IT company. The forecast was based on the company's billing data for 192 months of activity. The analysis of the data was based on the Cross Industry Standard Process for Data Mining approach and for the treatment; we used the Anaconda IPython and Pandas. We developed the prediction with three models using R: Exponential Smoothing (Holt-Winters), autoregressive integrated moving average (ARIMA) and artificial neural networks (ANN). The comparison of the performance of each of the methods shows that the model based on artificial neural networks has a greater accuracy in the prediction. These results need deepening the study to broaden the universe of the studied contexts. However, the simplicity in the application of the artificial neural networks model makes possible its use in computer applications without specific knowledge, giving a reliable instrument that allows the supporting decision-making by managers.
- BUSINESS INTELLIGENCE IN THE TOOL MANAGEMENT USED BY THE CUT AND CNC MACHINES OF THE ORNAMENTAL ROCKS INDUSTRYPublication . Martinho, Domingos; TERESO, MARCO; Sobreiro, Pedro; Pratas, AntonioIn the ornamental stone industry, improving production processes is a constant challenge for managers looking for solutions that improve the competitiveness of the companies. The management of cutting and computerized cut command machine tools is one of the areas where this improvement in management processes can have a positive effect on the competitiveness and productivity of companies. With this purpose, a model based on business intelligence methodologies was developed to systematize and automate the management process. The proposed model consists of the following layers: acquisition, extract, transform and load, storage and access and analysis. The acquisition layer consists of the interface with data available in various formats. The extract, transform and load process aims to extract data from these repositories and load them into a data warehouse. While the access and analysis phase is based on the use of software tools with graphical user interface with advanced analysis reporting features. The technology infrastructure is supported by the open source Tibco Jaspersoft Community Edition software package, which provides tools for the practical implementation of the defined model. With this work, it is hoped to implement the defined business intelligence model thus giving answers to the problems identified by providing information for the decision-making that corresponds to the needs of the managers.
- PREVISÃO DO TEMPO DE PERMANÊNCIA EM DOIS LARES DO DISTRITO DE SANTARÉMPublication . Sobreiro, Pedro; Martinho, Domingos; Pratas, Antonio; TERESO, MARCOA previsão do tempo de permanência por parte dos utentes de lares, baseada na análise dos dados existentes, constitui um indicador da maior importância para os gestores deste tipo de equipamentos ajudando-os a planear os recursos necessários. Neste estudo desenvolveu-se uma análise para determinar o tempo de permanência de um idoso num lar e quais são as variáveis que influenciam esse tempo de permanência. Os dados foram analisados tendo por base o modelo de regressão de Cox onde se verificou que as variáveis faturação acumulada, Idade e distância da residência ao lar têm impacto no tempo de permanência. Na aplicação do teste log-rank encontramos diferenças na sobrevivência nas variáveis faturação acumulada e idade. Os resultados obtidos podem ser úteis para avaliação do tempo que decorre até ao abandono e dessa forma permitir planear os recursos necessários para acompanhar os utentes dos lares.
- Predicting High-Value Customers in a Portuguese Wine CompanyPublication . Sobreiro, Pedro; Martinho, Domingos; Pratas, Antonio; Garcia-Alonso, Jose; Berrocal, JavierWine companies operate in a very competitive environment in which they must provide better-customised services and products to survive and gain advantage. The high customer turnover rate is a problem for these companies. This work aims to provide wine companies with new knowledge about customers that help to retain the existing ones. The study applies a collected dataset from a transaction database in a medium-sized ortuguese wine company to determinate: (1) customer lifetime value; (2) cluster customer value as output (customer loyalty). The measurement of the customer lifetime value (CLV) was analysed using the Pareto/NBD model and gamma-gamma model. Clustering techniques are employed to segment customers according to Recency, Frequency, and Monetary (RFM) values. Study findings show that exists three clusters with different interest to the marketing strategies, identifying the high-value customers, to target using marketing to increase their lifetime value effectively. The implications for the marketing strategy decisions is that using techniques based on the RFM model can make the most from data of customers and transactions databases and thus create sustainable advantages.
- A SLR on Customer Dropout PredictionPublication . Sobreiro, Pedro; Martinho, DomingosDropout prediction is a problem that is being addressed with machine learning algorithms; thus, appropriate approaches to address the dropout rate are needed. The selection of an algorithm to predict the dropout rate is only one problem to be addressed. Other aspects should also be considered, such as which features should be selected and how to measure accuracy while considering whether the features are appropriate according to the business context in which they are employed. To solve these questions, the goal of this paper is to develop a systematic literature review to evaluate the development of existing studies and to predict the dropout rate in contractual settings using machine learning to identify current trends and research opportunities. The results of this study identify trends in the use of machine learning algorithms in different business areas and in the adoption of machine learning algorithms, including which metrics are being adopted and what features are being applied. Finally, some research opportunities and gaps that could be explored in future research are presented.
- Hybrid Random Forest Survival Model to Predict Customer Membership DropoutPublication . Sobreiro, Pedro; Martinho, DomingosDropout prediction is a problem that must be addressed in various organizations, as retaining customers is generally more profitable than attracting them. Existing approaches address the problem considering a dependent variable representing dropout or non-dropout, without considering the dynamic perspetive that the dropout risk changes over time. To solve this problem, we explore the use of random survival forests combined with clusters, in order to evaluate whether the prediction performance improves. The model performance was determined using the concordance probability, Brier Score and the error in the prediction considering 5200 customers of a Health Club. Our results show that the prediction performance in the survival models increased substantially in the models using clusters rather than that without clusters, with a statistically significant difference between the models. The model using a hybrid approach improved the accuracy of the survival model, providing support to develop countermeasures considering the period in which dropout is likely to occur.
- Teaching Sentiment in Emergency Online Learning—A Conceptual ModelPublication . Ricardo Vardasca, PhD, ASIS, FRPS; Martinho, Domingos; Sobreiro, PedroDue to the COVID-19 pandemic, higher education institutions with a face-to-face model have found themselves in the contingency of migrating to online learning. This study explores the perspective of all the lecturers at a Portuguese private higher education institution who were invited to participate, regardless of their research area, in this questionnaire. It aims to propose and test a conceptual model that combines attitudes, preferred activities, and technological experience with the sentiment about the impact of this experience on students’ learning process, on their teaching activity, and on the strategy of higher education institutions. An online questionnaire was conducted to 65 lecturers engaging in emergency online lecturing. The obtained results showed that lecturers reveal a positive attitude towards online lecturing, tend to prefer activities in which they feel most comfortable in face-to-face lecturing, and consider having technological experience useful for online activities. Lecturers have a positive sentiment about the impact of online learning on students’ learning, their faculty career, and the strategy of higher education institutions. The proposed conceptual model test shows that the model has well-fitting conditions. The results confirm the hypotheses formulated: namely, the predictive effect of attitude, preferred activities, and technological experience on sentiment. Faculty engagement in emergency online lecturing shows that the members are available to participate in the changing process, and the proposed conceptual model can be used to assess this readiness.
- Performance Evaluation of Information Systems in Portuguese MunicipalitiesPublication . Pratas, Antonio; Sobreiro, Pedro; Martinho, DomingosThe success of organizations increasingly depends on the performance of their information systems, and public bodies are no exception. If the service provided to all of us, as citizens, depends on this performance, then it is essential that we can evaluate it. This research evaluates the performance of information sys tems in the Portuguese municipalities, applying the DeLone and McLean model, which is based on the perceptions of the systems’ users. The research is based on a self-administered questionnaire to a representative sample of Portuguese munici palities employees. The six dimensions of the model were: system quality, infor mation quality, service quality, utilization, user satisfaction and impacts. A seventh dimension - demographics - and its relationship with the original model were also analyzed. The results show that DeLone and McLean model is suitable for the Por tuguese municipalities context, and it was concluded that the system quality and information quality dimensions have a major effect on user satisfaction and inten tion to use/use, respectively. It was also found that the user satisfaction dimension has a strong effect, and the intention to use/use dimension has a moderate effect on Impacts. These results provide relevant information for decision-making related to the evolution of information systems in Portuguese municipalities, showing the importance of investing in the systems quality and the quality of the information produced.