Abstract
In order to improve their competitive performance, airline companies often adopt as a strategy to establish arrangement between two or more organizations agreeing to cooperate on a substantial level. This strategy is often known as airline alliances. A paradigm to analyze the collective intelligence behavior which emerges from a group, as a strategic alliance, is the flocking behavior. Inspired by the Cucker and Smale algorithm (C-S) we propose a new version of the flocking behavior algorithm applied to airline alliances. Our goal is to understand the link between strategic alliances and flocks. For this new approach, metrics were obtained for the parameters of C-S algorithm, namely position, velocity and influence, where the latter uses cooperative games. Besides, reinforcement learning mechanisms have been explored. Some relevant outputs for airline alliances as the permanence rate and the growth rate were computed for each of the five configurations in analysis.

Abstract
In the past years, data has become increasingly fast and volatile, making the ability to track its evolution an highly significant part of the value extraction process. In this work we present a framework to monitor evolution of clusters and present its use on real world data. We develop a framework over a previous one by Oliveira and Gama from 2013. Its biggest contribution is the addition of the concept of control area. This area will create a region around the cluster where it is still possible to establish associations with clusters from other time points. It aims to expand the search scope for cluster associations while diminishing the number of false positives. Changes to the transition definitions and detection algorithm are also introduced to accommodate the existence of this area. We demonstrate this framework at work in a real world scenario testing it with a telecom industry dataset and make a detailed analysis of the obtained results.

Abstract

Abstract
Recommender systems represent user preferences for items that the user might be interested to view or purchase. These systems have become extremely common in electronic commerce, providing relevant suggestions and directing users towards those items that best meet their needs and preferences. Different techniques have been analysed including content-based, collaborative and hybrid approaches. The last one is used to improve performance prediction combining different recommender systems using the best features of each method, smoothing problems as cold-start. We evaluate our ensemble method using MovieLens dataset with promising results.

Abstract
Telecommunication companies are acknowledging the existing connection between customer satisfaction and company revenues. Customer churn in telecom refers to a customer that ceases his relationship with a company. Churn prediction in telecom has recently gained substantial interest of stakeholders, who noticed that retaining a customer is substantially cheaper that gaining a new one. This research compares six approaches using different algorithms that identify the clients who are closer to abandon their telecom provider. Those algorithms are: KNN, Naive Rayes, C4.5, Random Forest, AdaBoost and ANN. The use of real data provided by WeDo technologies extended the refinement time necessary, but ensured that the developed algorithm and model can be applied to real world situations. The models are evaluated according to three criteria: are under curve, sensitivity and specificity, with special weight to the first two criteria. The Random Forest algorithm proved to be the most adequate in all the test cases.

Abstract
Built-in sensors in most modern smartphones open multiple opportunities for novel context-aware applications. Although the Human Activity Recognition field seized such opportunity, many challenges are yet to be addressed, such as the differences in movement by people doing the same activities. This paper exposes empirical research on Online Semi-supervised Learning (OSSL), an under-explored incremental approach capable of adapting the classification model to the user by continuously updating it as data from the user's own input signals arrives. Ultimately, we achieved an average accuracy increase of 0.18 percentage points (PP) resulting in a 82.76% accuracy model with Naive Bayes, 0.14 PP accuracy increase resulting in a 83.03% accuracy model with a Democratic Ensemble, and 0.08 PP accuracy increase resulting in a 84.63% accuracy model with a Confidence Ensemble. These models could detect 3 stationary activities, 3 active activities, and all transitions between the stationary activities, totaling 12 distinct activities.