Abstract
Analytic approaches to combine interpretable models, although presented in different contexts, can be generalized to highlight the components that can be specialized. We propose a framework that structures the combination process, formalizes the problems that can be solved in alternative ways and evaluates the combined models based on their predictive ability to replace the base ones, without loss of interpretability. The framework is illustrated with a case study using data from the University of Porto, Portugal, where experiments were carried out. The results show that grouping base models by scientific areas, ordering by the number of variables and intersecting their underlying rules creates conditions for the combined models to outperform them. © 2019, Springer Nature Switzerland AG.

Abstract
Knowledge generalization of ruled-based models, such as decision trees or decision rules, have emerged from different backgrounds. This particular kind of models, given their interpretability, offer several possibilities to be combined. Despite each distinct context, common patterns have emerged revealing the systemic nature of the problem. In this paper, we look at the problem of generalizing the knowledge contained in a set of models as a process formalizing the operations that can be addressed in alternative ways. We also include a set-up to evaluate gen-eralized models based on their ability to replace the base ones from a predictive performance perspective, without loss of interpretability.

Abstract
Forecasting is an important task across several domains. Its generalised interest is related to the uncertainty and complex evolving structure of time series. Forecasting methods are typically designed to cope with temporal dependencies among observations, but it is widely accepted that none is universally applicable. Therefore, a common solution to these tasks is to combine the opinion of a diverse set of forecasts. In this paper we present an approach based on arbitrating, in which several forecasting models are dynamically combined to obtain predictions. Arbitrating is a metalearning approach that combines the output of experts according to predictions of the loss that they will incur. We present an approach for retrieving out-of-bag predictions that significantly improves its data efficiency. Finally, since diversity is a fundamental component in ensemble methods, we propose a method for explicitly handling the inter-dependence between experts when aggregating their predictions. Results from extensive empirical experiments provide evidence of the method's competitiveness relative to state of the art approaches. The proposed method is publicly available in a software package.

Abstract
Chatbots have been used in business contexts as a new way of communicating with customers. They use natural language to interact with the customers, whether while offering products and services, or in the support of a specific task. In this context, an important and challenging task is to assess the effectiveness of the machine-to-human interaction, according to business’ goals. Although several analytic tools have been proposed to analyze the user interactions with chatbot systems, to the best of our knowledge they do not consider user-defined criteria, focusing on metrics of engagement and retention of the system as a whole. For this reason, we propose the KnowBots tool, which can be used to discover relevant patterns in the dialogues of chatbots, by considering specific business goals. Given the non-trivial structure of dialogues and the possibly large number of conversational records, we combined sequential pattern mining and subgroup discovery techniques to identify patterns of usage. Moreover, a friendly user-interface was developed to present the results and to allow their detailed analysis. Thus, it may serve as an alternative decision support tool for business or any entity that makes use of this type of interactions with their clients. © Springer Nature Switzerland AG 2019.

Abstract
Critical health events represent a relevant cause of mortality in intensive care units of hospitals, and their timely prediction has been gaining increasing attention. This problem is an instance of the more general predictive task of early anomaly detection in time series data. One of the most common approaches to solve this problem is to use standard classification methods. In this paper we propose a novel method that uses a layered learning architecture to solve early anomaly detection problems. One key contribution of our work is the idea of pre-conditional events, which denote arbitrary but computable relaxed versions of the event of interest. We leverage this idea to break the original problem into two layers, which we hypothesize are easier to solve. Focusing on critical health episodes, the results suggest that the proposed approach is advantageous relative to state of the art approaches for early anomaly detection. Although we focus on a particular case study, the proposed method is generalizable to other domains. © Springer Nature Switzerland AG 2019.

Abstract
Sensors are susceptible to failure when exposed to extreme conditions over long periods of time. Besides they can be affected by noise or electrical interference. Models (Machine Learning or others) obtained from these faulty and noisy sensors may be less reliable. In this paper, we propose a data augmentation approach for making neural networks more robust to missing and faulty sensor data. This approach is shown to be effective in a real life industrial application that uses data of various sensors to predict the wear of an automotive fuel-system component. Empirical results show that the proposed approach leads to more robust neural network in this particular application than existing methods.