Abstract Cost‐sensitive ensemble learning as a combination of two approaches, ensemble learning and cost‐sensitive learning, enables generation of cost‐sensitive tree‐based ensemble models using the cost‐sensitive decision tree (CSDT) learning algorithm. In general, tree‐based models characterize nice graphical representation that can explain a model's decision‐making process. However, the depth of the tree and the number of base models in the ensemble can be a limiting factor in comprehending the model's decision for each sample. The CSDT models are widely used in finance (e.g., credit scoring and fraud detection) but lack effective explanation methods. We previously addressed this gap with cost‐sensitive tree Shapley Additive Explanation Method (CSTreeSHAP), a cost‐sensitive tree explanation method for the single‐tree CSDT model. Here, we extend the introduced methodology to cost‐sensitive ensemble models, particularly cost‐sensitive random forest models. The paper details the theoretical foundation and implementation details of CSTreeSHAP for both single CSDT and ensemble models. The usefulness of the proposed method is demonstrated by providing explanations for single and ensemble CSDT models trained on well‐known benchmark credit scoring datasets. Finally, we apply our methodology and analyze the stability of explanations for those models compared to the cost‐insensitive tree‐based models. Our analysis reveals statistically significant differences between SHAP values despite seemingly similar global feature importance plots of the models. This highlights the value of our methodology as a comprehensive tool for explaining CSDT models.

Quality of life (QoL) is one of the major issues for cancer patients. With the advent of medical databases containing large amounts of relevant QoL information it becomes possible to train predictive QoL models by machine learning (ML) techniques. However, the training of predictive QoL models poses several challenges mostly due to data privacy concerns and missing values in patient data. In this paper, we analyze several classification and regression ML models predicting QoL indicators for breast and prostate cancer patients. Two different approaches are employed for imputing missing values. The examined ML models are trained on datasets formed from two databases containing a large number of anonymized medical records of cancer patients from Sweden. Two learning scenarios are considered: centralized and federated learning. In the centralized learning scenario all patient data coming from different data sources is collected at a central location prior to model training. On the other hand, federated learning enables collective training of machine learning models without data sharing. The results of our experimental evaluation show that the predictive power of federated models is comparable to that of centrally trained models for short-term QoL predictions, whereas for long-term periods centralized models provide more accurate QoL predictions.

Human activities and behaviour in different domains are usually influenced by other people’s actions and opinion. Nowadays, it is evident that there is a growing research interest in sentiment analysis, evaluation and prediction. Content from web sources and social media is frequently used when people want to see others’ opinion about different things. Our research is focused on ML-based sentiment analysis of food services reviews data. The comparison of several regression models with regards to prediction of customer satisfaction of restaurant and food services is presented. The experimental data collected from food serving businesses located in Shanghai Lujiazui Commercial Zone includes keywords extracted from the customers’ written reviews. Additionally, the data are spatially labelled enabling to conduct separate analyses for different geographical regions. As a conclusion, the keywords extracted from the customer’s reviews were suitable for the prediction of three observed satisfaction criteria: food taste, service, and environment.

Since the very first occurrence, science was considered to be the most important aspect of human lives. With the growth of science and its fields the quality of live have grown as well. Over time science divided into two major categories: sciences that were focused to more practically help humanity - applied sciences, and those that helped development of other sciences and themselves in a theoretical manner - fundamental sciences. We give the answer to which scientific fields are the most favored today and whether they are a part of applied or theoretical sciences by analyzing citation network of journals from one part of the Web of Science database. We introduce a measure called the probability of field superiority for ranking scientific fields based on journal citation networks. The probability of field superiority of field A over field B indicates that the randomly selected journal from field A has a bigger PageRank value than randomly selected journal from field B. The probability of field superiority was calculated for each category in Web of Science. Ranking of scientific fields shows which science branches are more favored today, which of them tend to be more perspective and where investments go. Journal rankings obtained from the network show that multidisciplinary journals tend to have the highest rankings. By using breadth first search algorithm and Tarjan's algorithm it was shown that graph has a bow-tie structure that most of real-world directed networks tend to have. Walktrap algorithm detected 21 communities with pronounced community structure. Communities conduct mostly of multidisciplinary journals with mathematics as one of the fundamental sciences with the highest number of journals in each community.

Current Research Information Systems (CRISs) offer great opportunities for assessments of institutional research outputs and extraction of useful and actionable knowledge based on various data-analysis techniques. However, many of these opportunities have not been explored in depth, especially in c ulture-sensitive areas such as gender-based analysis of research productivity and collaboration. In this paper we present GERBER, a network-based methodology and accompanying tool for gender-based analysis of publication data stored in institutional CRISs. GERBER relies on statistically robust techniques applied on weighted co-authorship networks whose nodes are enriched with different types of researcher evaluation metrics. The functionality of GERBER is demonstrated on publication data stored in the institutional CRIS of the Faculty of Sciences, University of Novi Sad, Serbia. The obtained results show that GERBER enables institutional research managers and policy makers to detect gender inequalities and homophily in research productivity and collaboration. Finally, we discuss different possibilities to integrate GERBER with CRISs in order to facilitate continuous gender-based evaluation of researchers.