Recommendations for further work

As the whole study had a modular structure, at each and every module there is something more left to do. These partial conclusions can be found at the end of the respective parts (see sections 6, 10, 14).

Of course, as these modules were designed independently, they have a great degree of freedom and are extremely flexible. Which means that, as well as being able to work together, they can also be used separately, for different means, in different problems.

• For instance, the Time-Series prediction module (see sections 3, 4) can be used for forecasting purposes on different series than the ones used in the described application, and can be a part of a pure prediction tool.
• Therefore, for further work on the Time-Series module I suggest an extension of the system in order to find out what other problems it's applicable to. In constructing the new function I exploited the TS structure of SE events, therefore I expect that it can be useful for other problems which deal with TS forecasting.
• In the same way, the parallel structures shown and exploited for neural networks (see sections 7, 8) can be used on other problems which require neural network implementation, completely different from what was discussed in this thesis. As a matter of fact, in section 8 I have discussed a parallelization of more than the Stock Exchange application that was used in subsequent sections.
• Also, in section 7, I have discussed optimized configurations of Neural Networks that can be implemented on different parallel architectures. In this project I have obtained some optimizations even in a UNIX environment, but these ideas can work better by using threads, or on a PVM environment, for instance, or MPI, or even better on MPP architectures. I have drawn some guidelines in this direction for further research.
• Also, the Extraction Environment, the SSKEE, can be exploited in a different manner. Not only can it function for different fields in classroom teaching, that require working with data that are hard to represent strictly as a function or as a clear set of rules, but also, the aim of the tool can be changed, so that its ultimate purpose can differ from classroom teaching. For instance, it can be even used in prediction, to generate some approximate rules of the data behaviour, or generally, when data analysis is needed and a symbolic representation is required.
• Hybrid Learning (HL) Systems - able to exploit simultaneously theoretical and empirical data, would be more efficient than each of the Explanation-Based Learning (EBL) systems - that use only theoretical knowledge in symbolical form, or Empirical Learning (EL) systems - handling the knowledge in the empirical form, working separately.
• The EBL and EL approaches are complementary in many aspects, so they can mutually offset weaknesses and alleviate inherent problems. It is even hoped that HL Systems could assist the induction of scientific theories, helping discover salient features in the input data, the importance of which could be over-looked.
• For the SSKEE, further work could enhance this environment, by adding more modules for rule extraction.
• Also, different teaching fields can be added to the system.
• Finally, it would be useful to make this NNKEE environment available on the Internet, for an easy access for teachers everywhere.
• For the Educational field, higher order (meta-) knowledge is important. On the present example, such knowledge processing as suggested does not yet exist. This item can also belong to future research.