Coleção de Artigos Acadêmicos
URI permanente para esta coleçãohttps://repositorio.insper.edu.br/handle/11224/3227
Navegar
10 resultados
Resultados da Pesquisa
Artigo Científico Validação de modelos de machine learning por experimentos estatísticos de campo(2024) Toaldo, Alexsandro; Vallim Filho, Arnaldo Rabello de Aguiar; Oyadomari, José Carlos Tiomatsu; Mendonça Neto, Octavio Ribeiro deObjetivo – Este artigo apresenta uma aplicação prática com o desenvolvimento de um experimento estatístico de campo em uma indústria de latas premium de alumínio nos Estados Unidos, visando validar estatisticamente resultados de modelos de machine learning (ML), previamente construídos. Metodologia: Usou-se conceitos de pesquisa intervencionista, que envolve experimentos de campo onde pesquisador e organização anfitriã atuam em conjunto buscando experimentar no sistema em estudo, e por meio da observação gerar conhecimento. Originalidade/Relevância: Sobre originalidade, não é frequente na literatura modelos de ML validados por experimento planejado de campo, seguido de análise estatística rigorosa. E a relevância da proposta se deve à sua contribuição para a literatura e pelas possibilidades de replicações do estudo em escala maior, na própria empresa ou em qualquer outra com desafios similares. Principais Resultados: Em fase anterior do estudo modelos de ML identificaram as variáveis de maior impacto em ineficiências (geração de sucata) em um processo de produção de latas de alumínio. Essas variáveis foram validadas nesta fase do estudo, através de experimento estatístico de campo, confirmando a significância estatística dos resultados do modelo de ML. Contribuições Teóricas e Práticas: A pesquisa contribui em termos práticos e científicos, pois a validação estatística de modelos de ML por experimentos planejados de campo é uma contribuição para a literatura de ciência aplicada, além de usas possibilidades práticas. Da mesma forma, apesar de amplamente utilizadas em diferentes áreas, pesquisas de cunho intervencionista ainda apresentam lacuna importante nas ciências sociais aplicadas, principalmente na gestão de processos industriais.Artigo Científico How Reliable Are The Screening Tools As A Triage Element For The Application Of The Global Leadership Initiative On Malnutrition (Glim)? Prospective Multicenter Observational Study(2023) Lopes, G.G.; Piovacari, S. M. F.; Moraes, J. R.; Santos, H. B. C.; Rakovicius, A. K. Z.; ANDRE FILIPE DE MORAES BATISTA; Pereira, A.J.Rationale: The international GLIM guideline recommends on the use of nutritional screening for the diagnosis of hospital malnutrition. However, as clinical outcomes were not included in the original validation of these instruments and their sensitivity (true positive rate) is unknown, it is not possible to report a chance of a truly at nutritional risk patient not being identified by screening and consequently not being evaluated by the GLIM. Methods: Multicenter, prospective observational trial in 3 Brazilian tertiary hospitals, carried out between December/21 to February/22. A convenience sample was used based on patients with expected length of stay at hospital longer than 48 hours. Pregnant women, under 18 years old, palliative care, lymphedema and muscle atrophy of neurological causes patients were excluded. NRS 2002, MNA-SF and ESPEN 2019 tools were applied to the specific populations (following international guidelines) by trained and validated Dietitians. Hospital mortality data were extracted from the local electronic medical records. Results: 676 patients were included, 54% male, 90% white with a mean age of 63 years (SD: ±21) and BMI of 27.50 kg/m2 (SD: ±4.72), hospitalized in the wards (68%). The most used nutritional screening was NRS 2002 (52%). In the sample, 39% were at nutritional risk, of these 5.6% died. An overview of nutritional screening tools’ performance are shown in the Table. In addition, accuracy found was 0.59 and area under the curve was 0,69 for predicting in-hospital deaths.Artigo Científico Data-driven decision making for the screening of cognitive impairment in primary care: a machine learning approach using data from the ELSA-Brasil study(2023) Szlejf, C.; ANDRE FILIPE DE MORAES BATISTA; Bertola, L.; Lotufo, P.A.; Benseñor, I.M.; Chiavegatto Filho, A.D.P.; Suemoto, C.K.The systematic assessment of cognitive performance of older people without cognitive complaints is controversial and unfeasible. Identifying individuals at higher risk of cognitive impairment could optimize resource allocation. We aimed to develop and test machine learning models to predict cognitive impairment using variables obtainable in primary care settings. In this cross-sectional study, we included 8,291 participants of the baseline assessment of the ELSA-Brasil study, who were aged between 50 and 74 years and were free of dementia. Cognitive performance was assessed with a neuropsychological battery and cognitive impairment was defined as global cognitive z-score below 2 standard deviations. Variables used as input to the prediction models included demographics, social determinants, clinical conditions, family history, lifestyle, and laboratory tests. We developed machine learning models using logistic regression, neural networks, and gradient boosted trees. Participants’ mean age was 58.3±6.2 years, 55% were female. Cognitive impairment was present in 328 individuals (4%). Machine learning algorithms presented fair to good discrimination (areas under the ROC curve between 0.801 and 0.873). Extreme Gradient Boosting presented the highest discrimination, high specificity (97%), and negative predictive value (97%). Seventy-six percent of the individuals with cognitive impairment were included among the highest ranked individuals by this algorithm. In conclusion, we developed and tested a machine learning model to predict cognitive impairment based on primary care data that presented good discrimination and high specificity. These characteristics could support the detection of patients who would not benefit from cognitive assessment, facilitating the allocation of human and economic resources.Artigo Científico Alpha-maxmin as an aggregation of two selve(2024) Chateauneuf, Alain; JOSÉ HELENO FARO; Tallon, Jean-Marc; Vergopoulos, VassiliThis paper offers a novel perspective on the -maxmin model, taking its components as originating from distinct selves within the decision maker. Drawing from the notion of multiple selves prevalent in inter-temporal decision-making contexts, we present an aggregation approach where each self possesses its own preference relation. Contrary to existing interpretations, these selves are not merely a means to interpret the decision maker’s overall utility function but are considered as primitives. Through consistency requirements, we derive an -maxmin representation as an outcome of a convex combination of the preferences of two distinct selves. We first explore a setting involving objective information and then move on to a fully subjective derivation.Artigo Científico Deep learning models for inflation forecasting(2023) Theoharidis, Alexandre Fernandes; DIOGO ABRY GUILLEN; HEDIBERT FREITAS LOPES; Hosszejni, DarjusWe propose a hybrid deep learning model that merges Variational Autoencoders and Convolutional LSTM Networks (VAE-ConvLSTM) to forecast inflation. Using a public macroeconomic database that comprises 134 monthly US time series from January 1978 to December 2019, the proposed model is compared against several popular econometric and machine learning benchmarks, including Ridge regression, LASSO regression, Random Forests, Bayesian methods, VECM, and multilayer perceptron. We find that VAE-ConvLSTM outperforms the competing models in terms of consistency and out-of-sample performance. The robustness of such conclusion is ensured via cross-validation and Monte-Carlo simulations using different training, validation, and test samples. Our results suggest that macroeconomic forecasting could take advantage of deep learning models when tackling nonlinearities and nonstationarity, potentially delivering superior performance in comparison to traditional econometric approaches based on linear, stationary models.- When It Counts—Econometric Identification of the Basic Factor Model Based on GLT Structures(2023) Frühwirth-Schnatter, Sylvia; Hosszejni, Darjus; HEDIBERT FREITAS LOPESDespite the popularity of factor models with simple loading matrices, little attention has been given to formally address the identifiability of these models beyond standard rotation-based identification such as the positive lower triangular (PLT) constraint. To fill this gap, we review the advantages of variance identification in simple factor analysis and introduce the generalized lower triangular (GLT) structures. We show that the GLT assumption is an improvement over PLT without compromise: GLT is also unique but, unlike PLT, a non-restrictive assumption. Furthermore, we provide a simple counting rule for variance identification under GLT structures, and we demonstrate that within this model class, the unknown number of common factors can be recovered in an exploratory factor analysis. Our methodology is illustrated for simulated data in the context of post-processing posterior draws in sparse Bayesian factor analysis.
Artigo Científico Probabilistic Nearest Neighbors Classification(2024) Fava, Bruno; PAULO CILAS MARQUES FILHO; HEDIBERT FREITAS LOPESAnalysis of the currently established Bayesian nearest neighbors classification model points to a connection between the computation of its normalizing constant and issues of NP-completeness. An alternative predictive model constructed by aggregating the predictive distributions of simpler nonlocal models is proposed, and analytic expressions for the normalizing constants of these nonlocal models are derived, ensuring polynomial time computation without approximations. Experiments with synthetic and real datasets showcase the predictive performance of the proposed predictive model.Artigo Científico Stochastic Volatility Models with Skewness Selection(2024) Martins, Igor; HEDIBERT FREITAS LOPESThis paper expands traditional stochastic volatility models by allowing for time-varying skewness without imposing it. While dynamic asymmetry may capture the likely direction of future asset returns, it comes at the risk of leading to overparameterization. Our proposed approach mitigates this concern by leveraging sparsity-inducing priors to automatically select the skewness parameter as dynamic, static or zero in a data-driven framework. We consider two empirical applications. First, in a bond yield application, dynamic skewness captures interest rate cycles of monetary easing and tightening and is partially explained by central banks’ mandates. In a currency modeling framework, our model indicates no skewness in the carry factor after accounting for stochastic volatility. This supports the idea of carry crashes resulting from volatility surges instead of dynamic skewness.Artigo Científico Decoupling Shrinkage and Selection in Gaussian Linear Factor Analysis(2024) Bolfarine, Henrique; Carvalho, Carlos M.; HEDIBERT FREITAS LOPES; Murray, Jared S.Factor analysis is a popular method for modeling dependence in multivariate data. However, determining the number of factors and obtaining a sparse orientation of the loadings are still major challenges. In this paper, we propose a decision-theoretic approach that brings to light the relationship between model fit, factor dimension, and sparse loadings. This relation is done through a summary of the information contained in the multivariate posterior. A two-step strategy is used in our method. First, given the posterior samples from the Bayesian factor analysis model, a series of point estimates with a decreasing number of factors and different levels of sparsity are recovered by minimizing an expected penalized loss function. Second, the degradation in model fit between the posterior of the full model and the recovered estimates is displayed in a summary. In this step, a criterion is proposed for selecting the factor model with the best trade-off between fit, sparseness, and factor dimension. The findings are illustrated through a simulation study and an application to personality data. We used different prior choices to show the flexibility of the proposed method.Artigo Científico Sparse Bayesian Factor Analysis When the Number of Factors Is Unknown(2024) Frühwirth-Schnatter, Sylvia; Hosszejni, Darjus; HEDIBERT FREITAS LOPESThere has been increased research interest in the subfield of sparse Bayesian factor analysis with shrinkage priors, which achieve additional sparsity beyond the natural parsimonity of factor models. In this spirit, we estimate the number of common factors in the widely applied sparse latent factor model with spike-and-slab priors on the factor loadings matrix. Our framework leads to a natural, efficient and simultaneous coupling of model estimation and selection on one hand and model identification and rank estimation (number of factors) on the other hand. More precisely, by embedding the unordered generalised lower trian gular loadings representation into overfitting sparse factor modelling, we obtain posterior summaries regarding factor loadings, common factors as well as the factor dimension via postprocessing draws from our efficient and customized Markov chain Monte Carlo scheme.