D4.2 The ML approach for the DST
This deliverable presents the specific use case application of the decision support tool (DST) functionality, as laid out in the D 4.1 deliverable of the project. In D 4.1 we introduced the design concepts and building blocks of a fully-fledged cloud DST. In the current deliverable we, principally, present its application against the SMEthod data set. As explained in good detail in D 4.1, we have deployed classification technology via machine learning (ML) and in particular neural network (NN) & Deep Learning (DL) approaches. Classification allows mapping the data collected by innovation stakeholders in certain classes, or segments, as they are referred to. Although the DST has been developed with ML/ NN classification in mind, it can, however also be used in more conventional statistical modelling. Thus, innovation stakeholders preferring to manage their data in a more traditional way may still find the DST useful. However, statistical modelling has not been implemented in the LEIMINTE platform; we have restricted to ML approaches as discussed above and as provided for in the Grant Agreement. In the development of the DST we had to embark much earlier than was initially provided for in the Grant Agreement. The development of LEIMINTE took overall one year to reach a stable and workable environment. In the course of this 1- year development, we had to use test data and schemata/ models for testing purposes. The SMEthod data collection was conducted in September 2018, whereas the segmentation analysis was concluded in July 2019. Also, instead of using completely arbitrary data and schemata we preferred to develop some- in coordination with innovation stakeholders- hoping that we would, in this way, come as close as possible to their aspirations. Thus, in this document, we will also review this key preparatory activity, before the introduction of the project data. In short, this document comprises two parts: • the LEIMINTE testing approach and development phase as well as the data and data models, this was based on. • the usage of the actual project data itself in order to carry out specific, selected, ML classification exercises within LEIMINTE.
Download the full report (PDF) here.