19 February 2021 in Statistical Analysis and Data Mining: An ASA Data Science Journal
The importance of providing explanations for predictions made by black‐box models has led to the development of explainer model methods such as LIME (local interpretable model‐agnostic explanations). LIME uses a surrogate model to explain the relationship between predictor variables and predictions from a black‐box model in a local region around a prediction of interest. However, the quality of the resulting explanations relies on how well the explainer model captures the black‐box model in a specified local region. Here we introduce three visual diagnostics to assess the quality of LIME explanations: (1) explanation scatterplots, (2) assessment metric plots, and (3) feature heatmaps. We apply the visual diagnostics to a forensics bullet matching dataset to show examples where LIME explanations depend on the tuning parameter values and the explainer model oversimplifies the black‐box model. Our examples raise concerns about claims made of LIME that are similar to other criticisms in the literature.
24 February 2025 in Geoscientific Model Development
Machine learning (ML) models are commonly used to generate predictions, but these models can also support the discovery of new science. Generating accurate predictions necessitates that a model captures the structure of the underlying data. If the structure is properly extracted, ML could be a useful exploratory and evidential tool. In this paper, we present a case study that demonstrates the use of ML for exploratory data analysis (EDA) in the climate space. We apply the ML explainability method of spatiotemporal zeroed feature importance (stZFI) to understand how climate-variable associations evolve over space and time. Our analyses focus on data from ensembles of Earth system models (ESMs) which provide data on different climate states and conditions. We elect to work with ESM ensembles since they allow us to compare feature importance across alternative scenarios not available with observed data. The ensembles also account for natural variability so that we can distinguish between signal and noise due to natural climate variability when computing feature importance. The use of perturbed initial condition ensembles introduces variability mimicking the natural variability in the atmosphere; thus the signals emerging using feature importance (FI) can be evaluated against the natural variability in the climate system. For our analyses, we consider the 1991 volcanic eruption of Mount Pinatubo, which was a large stratospheric aerosol injection. We explore the climate pathway associated with the eruption from aerosols to radiation to temperature at both the near-surface and stratospheric levels. In addition to applying the method to data generated from two different ESMs, we apply stZFI to reanalysis data to compare the associations identified by stZFI. We show how stZFI tracks the importance of aerosol optical depth over time on forecasting temperatures. This case study illustrates usefulness of an ML tool (stZFI) for EDA on a well-studied climate exemplar.
23 March 2021 in North American Journal of Fisheries Management
Three species of invasive carp, Grass Carp (Ctenopharyngodon idella), Silver Carp (Hypophthalmichthys molitrix), and Bighead Carp (H. nobilis), are rapidly spreading throughout North America. Monitoring reproduction can help determine establishment in new areas but is difficult due to challenges associated with identifying fish eggs. Recently, random forest models provided accurate identification of eggs based on morphological traits, but the models have not been validated using independent data. Our objective was to evaluate the predictive performance of egg identification models developed by Camacho et al. (2019) for classifying invasive carp eggs using an independent dataset. When invasive carp were grouped as one category, predictive accuracy was high at the family (89%), genus (90%), species (91%), and species with reduced predictor variables (94%) levels. Invasive carp predictive accuracy decreased when only considering observations from newly sampled locations [family (9%), genus (22%), species (30%), and species with reduced predictor variables (70%)], suggesting potential differences in egg characteristics among locations. Random forest models using a combination of previous and new data resulted in high predictive accuracy for invasive carp (96% to 98%) when invasive carp were grouped as one class for all models at the family, genus, and species levels. The two most influential predictor variables were average membrane diameter and average embryo diameter, where the probability of predicting an invasive carp egg increased with these metrics. High predictive accuracy metrics suggest that these trained and validated random forest models can be used to identify invasive carp eggs based on morphometric variables. However, decreased performance at new locations suggests more research would be beneficial to determine the applicability of the models to a larger spatial region.
27 February 2023 in PeerJ
The collection of fish eggs is a commonly used technique for monitoring invasive carp. Genetic identification is the most trusted method for identifying fish eggs but is expensive and slow. Recent work suggests random forest models could provide an inexpensive method for identifying invasive carp eggs based on morphometric egg characteristics. While random forests provide accurate predictions, they do not produce a simple formula for obtaining new predictions. Instead, individuals must have knowledge of the R coding language, limiting the individuals who can use the random forests for resource management. We present WhoseEgg: a web-based point-and-click application that allows non-R users to access random forests via a point and click interface to rapidly identify fish eggs with an objective of detecting invasive carp (Bighead, Grass, and Silver Carp) in the Upper Mississippi River basin. This article provides an overview of WhoseEgg, an example application, and future research directions.
23 September 2024 in EGUsphere
Machine learning (ML) models are commonly used to generate predictions, but these models can also support the discovery of new science. Generating accurate predictions necessitates that a model captures the structure of the underlying data. If the structure is properly extracted, ML could be a useful exploratory and evidential tool. In this paper, we present a case study that demonstrates the use of ML for exploratory data analysis (EDA) in the climate space. We apply the ML explainability method of spatio-temporal zeroed feature importance (stZFI) to understand how climate variable associations evolve over space and time. Our analyses focus on data from ensembles of earth systems models (ESMs), which provide data on different climate states and conditions. We elect to work with ESM ensembles since they allow us to compare feature importance across alternative scenarios not available with observed data. The ensembles also account for natural variability, so we can distinguish between signal and noise due to natural climate variability when computing feature importance. For our analyses, we consider the 1991 volcanic eruption of Mount Pinatubo: a large stratospheric aerosol injection. We explore the climate pathway associated with the eruption from aerosols to radiation to temperature at both the near-surface and stratospheric levels. In addition to applying the method to data generated from two different ESMs, we apply stZFI to reanalysis data to compare the associations identified by stZFI. We show how stZFI tracks the importance of aerosol optical depth over time on forecasting temperatures. This case study illustrates usefulness of an ML tool (stZFI) for EDA on a well studied climate exemplar.