From the set of dimensionality reduction techniques, we focus on the feature selection, a possible approach to carry it out is the use of Logical Analysis of Inconsistent Data (LAID). Recently, studies have demonstrated its potential in solving this problem and highlighted its advantages as a robust methodology, easy to interpret, and additionally capable of dealing with inconsistent data. The same studies revealed processing times higher than desired for full use and recommended the execution of algorithms through parallel processing using high-performance computing (HPC). This work represents yet another contribution to this effort by addressing ways of storing data, solutions for parallelizing algorithms, configuring the HPC environment, and finally testing the National Infrastructure for Distributed Computing (INCD) that allowed us to draw the conclusions presented.

Data Science applied to Marketing has been a research interest due to competitive advantages in business. We have applied a systematic literature review between 2010 and 2020, reaching a total of 19 valid articles. After a deeper segmentation, 13 articles were selected for inclusion in the review comprising the period 2013-2020. On scientific production, the topic Data Science Applied to Marketing, in 2020, has a new subject of interest. The number of citations has been growing since 2015 and the findings revealed that marketing is recurring of a variety of data science methods, from micro-segmentation and realtime application to natural language processing. The impact is evident in digital advertising, micro-segmentation and micro-targeting, speed and performance, and real-time experimentation. The use cases of data analytics in marketing have used four methods with the highest potential to impact marketing approaches: Internet-of-Things, big data, artificial intelligence, and machine learning.

O financiamento coletivo, ou crowdfunding, é uma designação nova para uma prática ancestral, a cotização coletiva para a viabilização de um projeto. Como ati- vidade contemporânea, passa a ser organizada por plataformas digitais e permite que desconhecidos se organizem em torno de projetos que não teriam incentivos para ser re- alizados em outras circunstâncias. Buscamos identificar quais são os determinantes dos apoios em projetos publicados no Catarse, principal plataforma do tipo no Brasil. Para realizar esse estudo, analisamos a literatura existente a respeito do contexto tecnológico e social no qual surgem iniciativas no campo da Economia do Compartilhamento e da Gig economy, e dos estudos anteriores sobre crowdfunding para delimitar o escopo teórico. Posteriormente, coletamos informações a respeito de aproximadamente 16 mil projetos e 950 mil usuários cadastrados através de web scraping e analisamos os dados utilizando regressão binomial negativa em multinível. Como conclusão, percebemos que realizadores que planejam melhor suas estratégias de comunicação atraem mais apoios. Se mostraram variáveis significativas o número de atualizações, a utilização de outras plataformas de comunicação e o tamanho das descrições. Algumas características dos projetos também se mostraram significativas, como o número de recompensas, a duração da campanha e a modalidade no qual estão inseridas. Da mesma forma, algumas características dos realizadores se mostraram relevantes, como uma rede pessoal de contatos e experiências anteriores. Por fim, percebemos efeitos de geografia no número de apoios, como a den- sidade populacional e a concentração de renda dos municípios de onde se originam os projetos. Como contribuição ao campo da comunicação, este trabalho ambiciona avançar a compreensão a respeito das novas configurações sociais surgidas com a internet e propor uma reflexão a respeito de como as pesquisas em comunicação podem se apropriar de ferramentas de outras áreas para aprimorar seus próprios métodos.

zenPrice™ is a SaaS solution created by the company Ritain.io that collects, via web-scrapping, the prices of various products in the e-commerce market and then makes them available through a centralised platform to its customers, which are usually companies that also sell this type of products. The platform can be improved through the introduction of new algorithms and methods capable of better capturing patterns and important information in the data. After meetings with Ritain.io, three functionalities capable of producing relevant insights were identified: multi-day price forecast, one-day price change forecast and competitor profile identification. The objective of this work is the implementation and study of techniques and statistical models that can later serve as a basis for the development of those functionalities. To carry out the multi-day price forecast, the ARIMA and Prophet models from Facebook were used, the latter having achieved the desired result when used in a multivariate approach, which led to the conclusion that using only the prices of the previous days of a product to predict the future prices of that same product is insufficient. Predicting a price change is a much simpler problem than predicting prices and, as this is a discrete and not continuous problem, different models have been used, such as Markov Chains and LSTM. Finally, to identify a competitor’s profile, which, in a simplified way, involves identifying time-varying lag and lead relationships between time series, an analysis of the coefficients of linear regression models was performed. Other approaches, such as Dynamic Time Warping and cross-correlation, were also discussed. An overview of the research work and the Python code used to generate the results presented in this dissertation are available at Github (github.com/ieeta-pt/zenPriceTSA).

Competition in the commercial market is increasingly fierce. In general, companies want to be able to predict the behavior of the market, in order to adapt the price of their products so that they can remain active and competitive in this market. It is well-known that the market is volatile and constantly changing, so every day there may be a change in the value of products. This is what makes the theme of this dissertation relevant, its goal is to find patterns of interest in the market, making it possible to provide valuable information to the company that will help decision making related to the prices of its products. This work was developed in partnership with the company Ritain.io, under the project zenPrice™ - Competitive intelligence SaaS, based on time series with the evolution of the price of a given product during a certain period of time, where the daily price is read and stored in an information repository. The identification of patterns in the market can be done through clustering procedures that allow finding groups of similar products taking into account the characteristics of the time series. Along this dissertation, features (or measures) are explored in the time, spectral and statistical domains. Furthermore, there is an analysis of the impact on the clustering process when different features are chosen. Since this work makes use of a high volume of data, it is convenient to consider a reduction in the dimensionality of the data before performing the clustering, in order to reduce the complexity of the problem and the need for computational resources. Therefore, this dissertation will focus on nonlinear methods for dimensionality reduction, more specifically the t-SNE, where the influence of different parameters is also studied. Using this method and the k-means clustering algorithm, 5 clusters were formed, where 3 of them are non-separable and 2 clusters are very cohesive and separable from the remaining 3.

This dissertation delves into predictive modeling for estimating the Length of Stay (LOS) of Intensive Care Unit (ICU) patients diagnosed with pneumonia. It employs a Deep Learning (DL)-based approach and compares it with Random Forest (RF) and Linear Regression (LR) models using the VertiCare database. Model evaluation employs various metrics including R-squared (R2), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and a novel Tolerance metric (accuracy of predictions within a range or interval of 3 days). The results begin with a baseline approach and proceed with incremental improvements. These enhancements include additional variables, codifications, patient group segmentation, and Box- Cox transformations. Notably, patient segmentation, involving exclusively patients diagnosed with pneumonia at hospital admission and excluding the patients with database prescription gaps, emerges as a significant catalyst for improved model performance. RF stood out achieving R2 and RMSE values around 0.62 and 5.7, respectively without Box-Cox transformation on LOS. With Box-Cox transformation, MAE and Tolerance got better results, yielding values of approximately 3.0 and 73.5%, respectively with RF. In the quest to optimize the DL model, the focus shifts to weight pruning aiming to reduce the complexity of the model while maximizing its predictive power. Two distinct pruning methodologies are explored, namely All-at-Once (involving sequential higher pruning percentages on the original model) and Iterative (entailing sequential pruning percentages applied to the previous pruned model. The Iterative pruning approach emerges as a superior option, achieving R2 values of around 0.65 and MAE values of approximately 3.0 for DL models with pruning percentages of up to 80%. This underscores the potential of Iterative pruning as a valuable optimization tool. Additionally, a novel feature selection technique is introduced. This method combines the insights from the Iterative pruning with feature importance based on the magnitude of the weights and reveals the optimal number of features as 12 out of 33 tested, combined with an 80% pruning threshold aiming a balance between model complexity and predictive accuracy. This work pioneers the application of these techniques in LOS prediction for ICU pneumonia patients using this database, offering insights for hospital resource management and decision-making.

This report is the result of an internship at the company Continental, as part of the Master's Degree in Data Science for Social Sciences, the aim of which is to demonstrate the possible uses of data in the company. Firstly, a brief description of the company is presented, in terms of its evolution, organisation and analysis of its financial data. The next stage is a detailed explanation of the two projects carried out, from the explanation of the software used to the interpretation of the data obtained from the analyses. This report, based on concrete cases, positions data analysis as an additional strategic variable to ensure decision-making and the competitiveness of companies.