No abstract

Smart learning ecosystems leverage on state-of-theart tools and technologies to help students learn better with Information Communication Technologies (ICT). The ubiquity, innovations and advancements of ICT have transformed pedagogies and approaches to content facilitation and delivery in higher education worldwide, the Pacific region being no exception. The paper essays a number of learning and support tools designed in-house or adopted (or outsourced) recently by a higher education institution in the Pacific contributing to the smart learning ecosystem. The institution has integrated these ICT driven tools to its academic and support programmes, and more recently the in-country science programmes introduced in its member countries. The strengths and challenges from the implementation of these new adaptive tools are highlighted with recommendations to the wider academic populace.

Science, Technology, Engineering and Mathematics (STEM) professionals play a key role in the development of an economy. STEM workers are critical thinkers as they contribute immensely by driving innovations.There is a high demand for professionals in the STEM fields but there is also a shortage of human resource in these areas. One way to reduce this problem is by identifying students who are at-risk of dropping out and then intervening with focused strategies that will ensure that these students remain in same the programme till graduation. Therefore, this research aims to use a data mining classification technique to identify students who are at-risk of dropping out from their Computing Science (CS) degree programmes. The Random Forest (RF) decision tree algorithm is used to learn patterns from historical data about first-year undergraduate CS students who are enrolled in a tertiary institute in the South Pacific. A number of factors are used which comprise of students demographic information, previous education background, financial information as well as data about students' academic interaction. Feature selection is performed to determine which factors have greater influence in students' decision in dropping out. Cross-validation techniques are used to ensure that the models are not over-fitted. Two models were built using a 5fold and 10-fold cross-validation and the results were compared using several measures of model performance. The results show that the factors corresponding to students' academic performance in a first-year programming course had the greatest impact student attrition in CS.

Student performance is a critical factor in determining a university's reputation because it has a negative effect on student retention. Students who do not perform well in a course are more likely to drop out from their programmes before graduating. Many students who enrol in Computing Science programmes struggle to find success because it is considered a difficult discipline. In this study, a sample of 918 observations were selected containing demographic and academic information about students enrolled in a first-year undergraduate Computing Science course at a university. Classification algorithms such as Decision Tree, Random Forest, Naïve Bayes and Support Vector Machine were used to build predictive models to determine whether a student will pass or fail the course. The results showed the Random Forest algorithms are capable of producing better predictive performance compared with traditional Decision Tree algorithms.

The breaking down of a particular problem through problem decomposition has enabled complex problems to be solved efficiently. The two major problem decomposition methods used in cooperative coevolution are synapse and neuron level. The combination of both the problem decomposition as a hybrid problem decomposition has been seen applied in time series prediction. The different problem decomposition methods applied at particular area of a network can share its strengths to solve the problem better, which forms the major motivation. In this paper, we are proposing a problem decomposition method that combines neuron and network level problem decompositions for Elman recurrent neural networks and applied to time series prediction. The results reveal that the proposed method has got better results in few datasets when compared to two popular standalone methods. The results are better in selected cases for proposed method when compared to several other approaches from the literature.