Suchergebnisse

Predicting academic success is essential in higher education because it is perceived as a critical driver for scientific and technological advancement and countries' economic and social development. This paper aims to retrieve the most relevant attributes for academic success by applying educational data mining (EDM) techniques to a Portuguese business school bachelor's historical data. We propose two predictive models to classify each student regarding academic success at enrolment and the end of the first academic year. We implemented a SEMMA methodology and tried several machine learning algorithms, including decision trees, KNN, neural networks, and SVM. The best classifier for academic success at the entry-level reached is a random forest with an accuracy of 69%. At the end of the first academic year, an MLP artificial neural network's best performance was achieved with an accuracy of 85%. The main findings show that at enrolment or the end of the first year, the grades and, thus, the student's previous education and engagement with the school environment are decisive in achieving academic success. Doi: 10.28991/ESJ-2023-SIED2-013 Full Text: PDF

Regional innovation performance is an important indicator for decision-making regarding the implementation of policies intended to support innovation. However, patterns in regional innovation structures are becoming increasingly diverse, complex and nonlinear. To address these issues, this study aims to develop a model based on a multi-output neural network. Both intra-and inter-regional determinants of innovation performance are empirically investigated using data from the 4th and 5th Community Innovation Surveys of NUTS 2 (Nomenclature of Territorial Units for Statistics) regions. The results suggest that specific innovation strategies must be developed based on the current state of input attributes in the region. Thus, it is possible to develop appropriate strategies and targeted interventions to improve regional innovation performance. We demonstrate that support of entrepreneurship is an effective instrument of innovation policy. We also provide empirical support that both business and government R&D activity have a sigmoidal effect, implying that the most effective R&D support should be directed to regions with below-average and average R&D activity. We further show that the multi-output neural network outperforms traditional statistical and machine learning regression models. In general, therefore, it seems that the proposed model can effectively reflect both the multiple-output nature of innovation performance and the interdependency of the output attributes. ; Regionální inovační výkonnost je důležitým ukazatelem pro rozhodování o provádění politik určených na podporu inovací. Vzory v regionálních inovačních strukturách se však stále více různí, jsou složité a nelineární. K řešení těchto problémů se tato studie zaměřuje na vývoj modelu založeného na neuronové síti s více výstupy. Empiricky se zkoumají jak vnitro- tak i mezi-regionální determinanty inovační výkonnosti s využitím dat ze 4. a 5. CIS pro NUTS 2 regiony. Výsledky naznačují, že je nutné vyvinout specifické inovační strategie založené na současném stavu vstupních atributů v regionu. Je tak možné vytvořit vhodné strategie a zásahy s cílem zlepšit regionální inovační výkonnost. Ukazujeme, že podpora podnikání je účinným nástrojem inovační politiky. Dále poskytujeme empirickou podporu, že jak podniková, tak vládní činnost v oblasti výzkumu a vývoje má sigmoidální efekt, což znamená, že nejúčinnější podpora výzkumu a vývoje by měla být zaměřena na regiony s podprůměrnou a průměrnou aktivitou v oblasti výzkumu a vývoje. Dále prokazujeme, že multi-výstupní neuronová síť překonává tradiční regresní modely a strojové učení. Obecně se proto zdá, že navrhovaný model může účinně odrážet jak charakter více výstupů inovačního výkonu, tak vzájemnou závislost výstupních atributů.

Hajek, P., & Henriques, R. (2017). Modelling innovation performance of European regions using multi-output neural networks. PLoS ONE, 12(10), [e0185755]. https://doi.org/10.1371/journal.pone.0185755 ; Regional innovation performance is an important indicator for decision-making regarding the implementation of policies intended to support innovation. However, patterns in regional innovation structures are becoming increasingly diverse, complex and nonlinear. To address these issues, this study aims to develop a model based on a multi-output neural network. Both intra- and inter-regional determinants of innovation performance are empirically investigated using data from the 4th and 5th Community Innovation Surveys of NUTS 2 (Nomenclature of Territorial Units for Statistics) regions. The results suggest that specific innovation strategies must be developed based on the current state of input attributes in the region. Thus, it is possible to develop appropriate strategies and targeted interventions to improve regional innovation performance. We demonstrate that support of entrepreneurship is an effective instrument of innovation policy. We also provide empirical support that both business and government R&D activity have a sigmoidal effect, implying that the most effective R&D support should be directed to regions with below-average and average R&D activity. We further show that the multi-output neural network outperforms traditional statistical and machine learning regression models. In general, therefore, it seems that the proposed model can effectively reflect both the multiple-output nature of innovation performance and the interdependency of the output attributes. © 2017 Hajek, Henriques. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. ; publishersversion ; published

Duarte, A., Henriques, R., & Ribeiro, S. (2019). Use of different optimization algorithms to define service areas of police stations in Portugal . In Evidence-based territorial policymaking: formulation, implementation and evaluation of policy: 26th APDR Congress Proceedings (pp. 108-115). Associacao Portuguesa para o Desenvolvimento Regional (APDR). ; Security is considered a fundamental right in democratic societies. Nowadays security forces reorganization depends on the population distribution and dynamics. The spatial disposition of security forces and the corresponding adequacy of resources benefit from the use of advanced spatial analysis tools. In order to provide better police service, it is necessary to divide the territory into areas or contiguous geographic clusters, which criteria may be having the same population, a similar crime rate, or balanced agent workloads. The process of grouping small geographic areas in service areas is called districting or zone design. The main objective of this work is the spatial optimization of the distribution of security forces considering the spatial disposition of the population in Setubal administrative district. In this paper, a genetic algorithm metaheuristic that addresses the police redistricting problem is proposed. The process of grouping areas into service areas was subject to constraints of population equality, contiguity and compactness. Data analysis was done with the hot spot analysis to study the population and their distribution at freguesia level. A genetic algorithm was implemented to create service areas (also known r-districs) and experimental analyses were performed with the population data and vulnerable groups. The simulations were made on 55 freguesias for 28 service areas. We compared the results with the Automatic Zoning Procedure - Simulated Annealing (AZP-SA). The population concentrates in Setubal peninsula, denoting a great asymmetry on their geographical distribution. Experimental tests with the genetic algorithm show a large variation of a sum of the population differences in service areas. Genetic algorithm presented on average 30402 inhabitants per service area. In the other hand, AZP-SA average was 29980. As regards the group with more than 65 years, both algorithms present on average 5599 inhabitants. The lowest compactness values were obtained with AZPSA in all tests performed. The absence of disaggregated data such as the number of crimes, geographical position of criminal incidents, the number of agents in the service, the number of cars per district were found as a drawback in our analysis. However, it is possible to simulate districts only with population data. The difficulty to obtain equitable areas is due to the great asymmetry of the population. ; publishersversion ; published

Doctoral studies are changing worldwide, with growing concerns about doctoral graduates' employability and ability to develop relevant links with industrial challenges. The present study aims examine the impact of Design Thinking skills on PhD students on their future academic and professional performance. Drawing on 7 years of pedagogical experimentation, we conducted a mixed methods longitudinal study to investigate the perceptions of students who attended a two-day Design Thinking workshop. Two questionnaires, with a total of 40 items measuring the quality and course impact dimensions, were given to 415 and 41 students, respectively. Finally, 12 students were chosen for in-depth interviews to learn more about how they applied their newly acquired design thinking skills in their research and work. Our findings show that developing Design Thinking skills impacts the professional lives of students of all fields of knowledge, ages, and stages of their PhD. The primary outcomes mentioned are associated with increased creative confidence and collaboration abilities. This study focuses on relevant dimensions for designing and delivering Design Thinking skills within doctoral programmes, as well as the impact of design thinking on the quality of PhD education and student employability opportunities. Doi: 10.28991/ESJ-2023-SIED2-02 Full Text: PDF

Castelli, M., Dobreva, M., Henriques, R., & Vanneschi, L. (2020). Predicting Days on Market to Optimize Real Estate Sales Strategy. Complexity, 2020, 1-22. [4603190]. https://doi.org/10.1155/2020/4603190 --- This work was supported by national funds through the FCT (Fundacao para a Ciencia e a Tecnologia) by the projects GADgET (DSAIPA/DS/0022/2018), BINDER (PTDC/CCI-INF/29168/2017), and AICE (DSAIPA/DS/0113/2019). Mauro Castelli acknowledges the financial support from the Slovenian Research Agency (research core funding no. P5-0410). ; Irregularities and frauds are frequent in the real estate market in Bulgaria due to the substantial lack of rigorous legislation. For instance, agencies frequently publish unreal or unavailable apartment listings for a cheap price, as a method to attract the attention of unaware potential new customers. For this reason, systems able to identify unreal listings and improve the transparency of listings authenticity and availability are much on demand. Recent research has highlighted that the number of days a published listing remains online can have a strong correlation with the probability of a listing being unreal. For this reason, building an accurate predictive model for the number of days a published listing will be online can be very helpful to accomplish the task of identifying fake listings. In this paper, we investigate the use of four different machine learning algorithms for this task: Lasso, Ridge, Elastic Net, and Artificial Neural Networks. The results, obtained on a vast dataset made available by the Bulgarian company Homeheed, show the appropriateness of Lasso regression. ; publishersversion ; published

Irregularities and frauds are frequent in the real estate market in Bulgaria due to the substantial lack of rigorous legislation. For instance, agencies frequently publish unreal or unavailable apartment listings for a cheap price, as a method to attract the attention of unaware potential new customers. For this reason, systems able to identify unreal listings and improve the transparency of listings authenticity and availability are much on demand. Recent research has highlighted that the number of days a published listing remains online can have a strong correlation with the probability of a listing being unreal. For this reason, building an accurate predictive model for the number of days a published listing will be online can be very helpful to accomplish the task of identifying fake listings. In this paper, we investigate the use of four different machine learning algorithms for this task: Lasso, Ridge, Elastic Net, and Artificial Neural Networks. The results, obtained on a vast dataset made available by the Bulgarian company Homeheed, show the appropriateness of Lasso regression.

Victorino, G., Bandeira, R., Painho, M., Henriques, R., & Coelho, P. S. (2022). Rethinking the Campus Experience in a Post-COVID World: A Multi-Stakeholder Design Thinking Experiment. Sustainability (Switzerland), 14(13), 1-13. [7655]. https://doi.org/10.3390/su14137655 ; In a post-COVID world, rethinking the campus experience is critical for defining new pedagogical strategies. As higher education moves toward more student-centred action learning models, university leaders should engage in democratic design methods that empower students and professors. Design thinking (DT) is a user-centred design approach that can aid in the creation of future learning environments. While DT has been used in innovative space design, we know little about how students, professors, and other community members can act as codesign partners. To understand their role in codesign and how their experiences are incorporated into new building design decisions, we need a conceptual model. To develop this model, we examined a case study of the evolutionary co-design process of a new building for a leading information management school in Europe. Using the concept of three phases of design thinking defined by Brown (2009): Inspire, Ideate, and Implement, we collaborated with a group of 50 design thinking students and more than 500 members of the community representing different stakeholders, to create new spaces and rethink the learning experience. Our discussion will centre on the creation of a participatory design thinking model that positions students as design partners alongside university decision makers. The findings conclude that, by applying design thinking methods, it was possible to unveil new dimensions of the success of future campuses that go beyond the building design. Creating meaningful learning spaces that inspire creativity and critical thinking requires an alignment between human centred design, organizational change management and new pedagogical strategies. ; publishersversion ; published

UID/CPO/04627/2013 PTDC/ATP­DEM/1538/2014 ; Demographic studies are a support instrument for decision-making in the field of security policies. Security is a public good, perceived as a critical political issue in the actual context of unstable balance between investment in internal security forces, and urgency of reducing criminality. SIM4SECURITY (PTDC/ATP-DEM/1538/2014) is a technological solution, based on a GIS model and demographic forecasting, which can improve the operational activity of the security forces. We aim to perform demographic disaggregated scenarios in order to optimize the ratio police/citizen, and assure the effectiveness of the security model based on new risks and threats, insecurity factors and the new necessities of local populations. Being a mutable system, the model will be able to monitor up until 2030 and take into account change of parameters, variables and factors that prove to influence dynamic population distributions scenarios. Os estudos demográficos são um instrumento de apoio à decisão no âmbito das políticas de segurança. A segurança é um bem público e representa uma questão política que exige a tomada de opções racionais entre investimento nas forças de segurança e a urgência de reduzir a criminalidade e os seus impactos. Como contributo para a resolução desta questão propomos a criação do SIM4SECURITY (PTDC/ATP-DEM/1538/2014), uma solução tecnológica baseada no desenvolvimento de um modelo SIG e na elaboração de cenários demográficos. Os resultados do exercício prospetivo permitirão otimizar a rácio polícia/cidadão e melhorar a eficácia do modelo de segurança, tendo em conta novos riscos e ameaças, novos fatores de insegurança e necessidades específicas adaptados às especificidades locais. O modelo é dinâmico e permite monitorizar até 2030 as mudanças de parâmetros, variáveis e fatores que venham a ser considerados influentes na distribuição da população portuguesa no território. ; publishersversion ; published