Sudha Ram’s research while affiliated with University of Arizona and other places

Incomplete data with blockwise missing patterns are commonly encountered in analytics, and solutions typically entail listwise deletion or imputation. However, as the proportion of missing values in input features increases, listwise or columnwise deletion leads to information loss, whereas imputation diminishes the integrity of the training data set. We present the blockwise reduced modeling (BRM) method for analyzing blockwise missing patterns, which adapts and improves on the notion of reduced modeling proposed by Friedman, Kohavi, and Yun in 1996 as lazy decision trees. In contrast to the original idea of reduced modeling of delaying model induction until a prediction is required, our method is significantly faster because it exploits the blockwise missing patterns to pretrain ensemble models that require minimum imputation of data. Models are pretrained over the overlapping subsets of an incomplete data set that contain only populated values. During prediction, each test instance is mapped to one of these models based on its feature-missing pattern. BRM can be applied to any supervised learning model for tabular data. We benchmark the predictive performance of BRM using simulations of blockwise missing patterns on three complete data sets from public repositories. Thereafter, we evaluate its utility on three data sets with actual blockwise missing patterns. We demonstrate that BRM is superior to most existing benchmarks in terms of predictive performance for linear and nonlinear models. It also scales well and is more reliable than existing benchmarks for making predictions with blockwise missing pattern data. Supplemental Material: The online appendix is available at https://doi.org/10.1287/ijds.2022.9016 .

We conducted a field study using multiple wearable devices on 231 federal office workers to assess the impact of the indoor environment on individual wellbeing. Past research has established that the workplace environment is closely tied to an individual’s wellbeing. Since sound is the most-reported environmental factor causing stress and discomfort, we focus on quantifying its association with physiological wellbeing. Physiological wellbeing is represented as a latent variable in an empirical Bayes model with heart rate variability measures—SDNN and normalized-HF as the observed outcomes and with exogenous factors including sound level as inputs. We find that an individual’s physiological wellbeing is optimal when sound level in the workplace is at 50 dBA. At lower (<50dBA) and higher (>50dBA) amplitude ranges, a 10 dBA increase in sound level is related to a 5.4% increase and 1.9% decrease in physiological wellbeing respectively. Age, body-mass-index, high blood pressure, anxiety, and computer use intensive work are person-level factors contributing to heterogeneity in the sound-wellbeing association.

The National Environmental Policy Act (NEPA) provides a regulatory decision-making process that requires U.S. federal agencies to assess the purpose and socio-environmental impacts of a proposed action before deciding to move forward with that action. The multiplicity of NEPA objectives, the complex tradeoffs embedded in the Environmental Impact Assessment (EIA), and the difficulties in accessing data have presented challenges for evaluating NEPA performance. Researchers have responded with a growing array of performance dimensions and specialized measurement approaches. In this paper, we advance a performance framework for EIAs that integrates several of these dimensions and provides a conceptually coherent approach to procedural and substantive performance. The framework articulates three procedural elements (use of science and analysis, nature of public participation, and management of EIA processes); and three substantive elements (quality of both the NEPA review and action decision, and both the accountability and efficiency of the NEPA review decision). Each element is further elaborated by specific functions with specific variables as a basis for future performance measurements. We use two hypothetical use cases, drawn from public land management and federal highway planning, to illustrate how the performance concepts from the framework can be operationalized and measured.

Wearables are an important source of big data as they provide real-time high-resolution data logs of health indicators of individuals. Higher-order associations between pairs of variables is common in wearables data. Representing higher-order association curves as piece-wise linear segments in a regression model makes them more interpretable. However, existing methods for identifying the change points for segmented modeling either overfit or have low external validity for wearables data containing repeated measures. Therefore, we propose a human-in-the-loop method for segmented modeling of higher-order pairwise associations between variables in wearables data. Our method uses the smooth function estimated by a generalized additive mixed model to allow the analyst to annotate change point estimates for a segmented mixed-effects model, and thereafter employs the Brent's constrained optimization procedure to fine-tuning the manually provided estimates. We validate our method using three real-world wearables datasets. Our method not only outperforms state-of-the-art modeling methods in terms of prediction performance but also provides more interpretable results. Our study contributes to health data science in terms of developing a new method for interpretable modeling of wearables data. Our analysis uncovers interesting insights on higher order associations for health researchers.

Conceptual modeling is important for developing databases that maintain the integrity and quality of stored information. However, classical conceptual models have often been assumed to work on well-maintained and high-quality data. With the advancement and expansion of data science, it is no longer the case. The need to model and store data has emerged for settings with lower data quality, which creates the need to update and augment conceptual models to represent lower-quality data. In this paper, we focus on the intersection between data completeness (an important aspect of data quality) and complex class semantics (where a complex class entity represents information that spans more than one simple class entity). We propose a new disaggregation construct to allow the modeling of incomplete information. We demonstrate the use of our disaggregation construct for diverse modeling problems and discuss the anomalies that could occur without this construct. We provide formal definitions and thorough comparisons between various types of complex constructs to guide future application and prove the unique interpretation of our newly proposed disaggregation construct.

Recent developments in big data technologies are revolutionizing the field of healthcare predictive analytics (HPA), enabling researchers to explore challenging problems using complex prediction models. Nevertheless, healthcare practitioners are reluctant to adopt those models as they are less transparent and accountable due to their black-box structure. We believe that instance-level, or local, explanations enhance patient safety and foster trust by enabling patient-level interpretations and medical knowledge discovery. Therefore, we propose the RObust Local EXplanations (ROLEX) method to develop robust, instance-level explanations for HPA models in this study. ROLEX adapts state-of-the-art methods and ameliorates their shortcomings in explaining individual-level predictions made by black-box machine learning models. Our analysis with a large real-world dataset related to a prevalent medical condition called fragility fracture and two publicly available healthcare datasets reveals that ROLEX outperforms widely accepted benchmark methods in terms of local faithfulness of explanations. In addition, ROLEX is more robust since it does not rely on extensive hyperparameter tuning or heuristic algorithms. Explanations generated by ROLEX, along with the prototype user interface presented in this study, have the potential to promote personalized care and precision medicine by providing patient-level interpretations and novel insights. We discuss the theoretical implications of our study in healthcare, big data, and design science.

We conducted a field study using multiple wearable devices on 231 federal office workers to assess the impact of the indoor environment on individual wellbeing. Past research has established that the workplace environment is closely tied to an individual’s wellbeing. Since sound is the most-reported environmental factor causing stress and discomfort, we focus on quantifying its association with physiological wellbeing. Physiological wellbeing is represented as a latent variable in an empirical Bayes model with heart rate variability measures – SDNN and normalized-HF as the observed outcomes and with exogenous factors including sound level as inputs. We find that an individual’s physiological wellbeing is optimal when sound level in the workplace is at 50 dBA. At lower (<50dBA) and higher (>50dBA) amplitude ranges, a 10 dBA increase in sound level is related to a 5.4% increase and 1.9% decrease in physiological wellbeing respectively. Age, body-mass-index, high blood pressure, anxiety, and computer use intensive work are person-level factors contributing to heterogeneity in the sound-wellbeing association.

Timely and accurate prediction of human movement in urban areas offers instructive insights into transportation management, public safety, and location-based services, to name a few. Yet, modeling urban mobility is challenging and complex because of the spatiotemporal dynamics of movement behavior and the influence of exogenous factors such as weather, holidays, and local events. In this paper, we use bus transportation as a proxy to mine spatiotemporal travel patterns. We propose a deep-learning-based urban mobility prediction model that collectively forecasts passenger flows between pairs of city regions in an origin-destination (OD) matrix. We first process OD matrices in a convolutional neural network to capture spatial correlations. Intermediate results are reconstructed into three multivariate time series: hourly, daily, and weekly time series. Each time series is aggregated in a long short-term memory (LSTM) network with a novel attention mechanism to guide the aggregation. In addition, our model is context-aware by using contextual embeddings learned from exogenous factors. We dynamically merge results from LSTM components and context embeddings in a late fusion network to make a final prediction. The proposed model is implemented and evaluated using a large-scale transportation data set of more than 200 million bus trips with a suite of Big Data technologies developed for data processing. Through performance comparison, we show that our approach achieves sizable accuracy improvements in urban mobility prediction. Our work has major implications for efficient transportation system design and performance improvement. The proposed deep neural network structure is generally applicable for sequential graph data prediction.

ABSTRACT The prediction of fracture risk in osteoporotic patients has been a topic of interest for decades, and models have been developed for the accurate prediction of fracture, including the fracture risk assessment tool (FRAX). As machine‐learning methodologies have recently emerged as a potential model for medical prediction tools, we aimed to develop a novel fracture prediction model using machine‐learning methods in a prospective community‐based cohort. In this study, 2227 participants (1257 females) with a baseline bone mineral density (BMD) and trabecular bone score were enrolled from the Ansung cohort. The primary endpoint was the fragility fractures reported by patients or confirmed by X‐rays. We used 3 different models: CatBoost, support vector machine (SVM), and logistic regression. During a mean 7.5‐year follow‐up (range, 2.5 to 10 years), fragility fractures occurred in 537 (25.6%) of participants. In predicting total fragility fractures, the area under the curve (AUC) values of the CatBoost, SVM, and logistic regression models were 0.688, 0.500, and 0.614, respectively. The AUC value of CatBoost was significantly better than that of FRAX (0.663; p

With recent advancements in data analytics, healthcare predictive analytics (HPA) is garnering growing interest among practitioners and researchers. However, it is risky to blindly accept the results and users will not accept the HPA model if transparency is not guaranteed. To address this challenge, we propose the RObust Local EXplanations (ROLEX) method, which provides robust, instance-level explanations for any HPA model. The applicability of the ROLEX method is demonstrated using the fragility fracture prediction problem. Analysis with a large real-world dataset demonstrates that our method outperforms state-of-the-art methods in terms of local fidelity. The ROLEX method is applicable to various types of HPA problems beyond the fragility fracture problem. It is applicable to any type of supervised learning model and provides fine-grained explanations that can improve understanding of the phenomenon of interest. Finally, we discuss theoretical implications of our study in light of healthcare IS, big data, and design science.