Third special issue on knowledge discovery and business intelligence

Expert Systems were proposed in the mid 1970s (Arnott & Pervan, 2014) with the goal of building computerized systems that mimic human behavior to solve real-world tasks. Such systems were based on artificial intelligence (AI) techniques, typically by adopting explicit (human understandable) knowledge, extracted from domain experts (e.g. by using interviews) and that was stored in a knowledge base (Buchanan, 1986). In the last decades, the world as changed due to advances in information and communication technology (e.g. massive usage of computers and personal mobile devices, Internet and social media, usage of digital cameras and other sensors). In effect, we are now in the age of data, where a large portion of organizational, societal or personal activities is captured digitally (Mojsilovic, 2014). Following this change, ES have evolved to include data-driven models, either solely or complemented by expert-driven knowledge. Such change is reflected in the ES journal, which currently publishes several articles related with data analysis fields (e.g. analytics and business intelligence, data mining and knowledge discovery, big data and data science). In this special issue, we highlight two of data related terms: knowledge discovery (KD) and Business Intelligence (BI). KD is often used as a synonym of data mining, and it consists in an AI subfield that uses machine learning algorithms to extract high-level interesting knowledge from raw data (Fayyad et al. 1996). BI is a popular management term (Arnott & Pervan, 2014), and it represents several technologies (e.g. data warehouses, KD and dashboards) that store and process organizational data in order to support managerial decision making (Delen et al. 2014). The ‘Knowledge Discovery and Business Intelligence’ (KDBI) thematic track was proposed for the EPIA conference on AI in 2009 with the goal of promoting the interaction between the KD and BI areas. Since then, the track has been included in all EPIA biennial conferences. After 2011, the KDBI track has been associated with special ES journal issues, which included extended versions of the best KDBI papers. The first special issue was published in 2013, and it included the best KDBI 2011 track papers (Cortez & Santos, 2013), while the second special issue appeared in 2015, and it encompassed the best KDBI 2013 track papers (Cortez & Santos, 2015). This issue, entitled ‘Third special issue on Knowledge Discovery and Business Intelligence’ , contains recent KD and BI contributions that can be used in ES to produce a valuable impact in real-world applications. It includes extended versions of papers from the 4th KDBI thematic track, of the 17th EPIA conference on AI (EPIA 2015), held in Coimbra, Portugal. The track received 18 paper submissions and the authors of the best papers were invited to extend their works for this special issue. After two rounds of reviews, which included reviewers from the KDBI track and also ES journal, the best six papers were accepted, corresponding to an overall acceptance rate of 33%. Due to the interest in data-driven models, in the last years, there has been several interesting developments in the KDBI area. Despite this progress, there are still many challenges and opportunities. For instance, most KD algorithms were targeted for single label classification tasks and often these algorithms cannot deal adequately with label ranking, which is useful in several real-world applications (e.g. modeling user preferences). Also, the financial domain still rises many challenges: it is not clear what is the best approach (regression or classification) to forecast trading actions; and easy to interpret tools are needed to better disclose the relationships among price variations from distinct financial products. Moreover, most of the real-world data has a temporal dimension and there is still room for proposing specialized algorithms that use this dimension in the KD or BI process (e.g. time series retrieval, visual representation of temporal changes). Furthermore, the ‘Extract, Transform and Load’ (ETL) is a vital component of BI systems, but it often requires a substantial manual effort in terms of its design and implementation, even when there are several common ETL processes that are repeated through distinct BI projects. All these challenges and opportunities are addressed in the six papers accepted in this special issue, which we will briefly detail in the next section.