Dataset Information

Marginal Contribution Feature Importance - an Axiomatic Approach for Explaining Data.

ABSTRACT: In recent years, methods were proposed for assigning feature importance scores to measure the contribution of individual features. While in some cases the goal is to understand a specific model, in many cases the goal is to understand the contribution of certain properties (features) to a real-world phenomenon. Thus, a distinction has been made between feature importance scores that explain a model and scores that explain the data. When explaining the data, machine learning models are used as proxies in settings where conducting many real-world experiments is expensive or prohibited. While existing feature importance scores show great success in explaining models, we demonstrate their limitations when explaining the data, especially in the presence of correlations between features. Therefore, we develop a set of axioms to capture properties expected from a feature importance score when explaining data and prove that there exists only one score that satisfies all of them, the Marginal Contribution Feature Importance (MCI). We analyze the theoretical properties of this score function and demonstrate its merits empirically.

SUBMITTER: Catav A

PROVIDER: S-EPMC8460841 | biostudies-literature | 2021 Jul

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Marginal Contribution Feature Importance - an Axiomatic Approach for Explaining Data.

Catav Amnon A Fu Boyang B Zoabi Yazeed Y Weiss-Meilik Ahuva A Shomron Noam N Ernst Jason J Sankararaman Sriram S Gilad-Bachrach Ran R

Proceedings of machine learning research 20210701

In recent years, methods were proposed for assigning feature importance scores to measure the contribution of individual features. While in some cases the goal is to understand a specific model, in many cases the goal is to understand the contribution of certain properties (features) to a real-world phenomenon. Thus, a distinction has been made between feature importance scores that explain a <i>model</i> and scores that explain the <i>data</i>. When explaining the data, machine learning models ...[more]

PMID: 34568830

Similar Datasets

Project description:PurposeCurrent limitations in methodologies used throughout machine-learning to investigate feature importance in boosted tree modelling prevent the effective scaling to datasets with a large number of features, particularly when one is investigating both the magnitude and directionality of various features on the classification into a positive or negative class. This manuscript presents a novel methodology, "Hollow-tree Super" (HOTS), designed to resolve and visualize feature importance in boosted tree models involving a large number of features. Further, this methodology allows for accurate investigation of the directionality and magnitude various features have on classification and incorporates cross-validation to improve the accuracy and validity of the determined features of importance.MethodsUsing the Iris dataset, we first highlight the characteristics of HOTS by comparing it to other commonly used techniques for feature importance, including Gini Importance, Partial Dependence Plots, and Permutation Importance, and explain how HOTS resolves the weaknesses present in these three strategies for investigating feature importance. We then demonstrate how HOTS can be utilized in high dimensional spaces such as neuroscientific setting, by taking 60 Schizophrenic subjects from the publicly available SchizConnect database and applying the method to determine which regions of the brain were most important for the positive and negative classification of schizophrenia as determined by the positive and negative syndrome scale (PANSS).ResultsHOTS effectively replicated and supported the findings of feature importance for classification of the Iris dataset when compared to Gini importance, Partial Dependence Plots and Permutation importance, determining 'petal length' as the most important feature for positive and negative classification. When applied to the Schizconnect dataset, HOTS was able to resolve from 379 independent features, the top 10 most important features for classification, as well as their directionality for classification and magnitude compared to other features. Cross-validation supported that these same 10 features were consistently used in the decision-making process across multiple trees, and these features were localised primarily to the occipital and parietal cortices, commonly disturbed brain regions in those afflicted with Schizophrenia.ConclusionHOTS effectively overcomes previous challenges of identifying feature importance at scale, and can be utilized across a swathe of disciplines. As computational power and data quantity continues to expand, it is imperative that a methodology is developed that is able to handle the demands of working with large datasets that contain a large number of features. This approach represents a unique way to investigate both the directionality and magnitude of feature importance when working at scale within a boosted tree model that can be easily visualized within commonly used software.

Project description:A common two-tier structure for social networks is based on partitioning society into two parts, referred to as the elite and the periphery, where the "elite" is the relatively small but well-connected and highly influential group of powerful individuals around which the society is centered, and the "periphery" consists of the rest of society. It is observed that the relative sizes of economic and social elites in various societies appear to be continually declining. One possible explanation is that this is a natural social phenomenon, resembling Price's "square root" law for the fraction of good scientists in the scientific community. We try to assess the validity of this explanation by studying the elite-periphery structure via introducing a novel axiom-based model for representing and measuring the influence between the elite and the periphery. The model is accompanied by a set of axioms that capture the elite's dominance, robustness and density, as well as a compactness property. Relying on the model and the accompanying axioms, we are able to draw a number of insightful conclusions about the elite-periphery structure. In particular, we show that in social networks that respect our axioms, the size of a compact elite is sublinear in the network size. This agrees with Price's principle but appears to contradict the common belief that the elite size tends to a linear fraction of society (recently claimed to be around 1%). We propose a natural method to create partitions with nice properties, based on the key observation that an elite-periphery partition is at what we call a 'balance point', where the elite and the periphery maintain a balance of powers. Our method is based on setting the elite to be the k most influential nodes in the network and suggest the balance point as a tool for choosing k and therefore the size of the elite. When using nodes degrees to order the nodes, the resulting k-rich club at the balance point is the elite of a partition we refer to as the balanced edge-based partition. We accompany these findings with an empirical study on 32 real-world social networks, which provides evidence that balanced edge-based partitions which satisfying our axioms commonly exist.

Project description:BackgroundMalnutrition among women disproportionately exists across socioeconomic classes of Bangladesh. According to our knowledge, studies which attempted to identify determinants and their contributions to explain BMI-based nutritional gap between the poorest and the richest categories of Wealth Index are still scarce.ObjectivesTo identify the nutritional gap of women between the richest-poorest classes in Bangladesh, and to determine how much of this gap are attributed to differences in predictors and differences in coefficients.Study populationReproductive-aged (15-49 years) women of Bangladesh.Methods and proceduresWe utilized the latest round (2017-2018) data of the Bangladesh Demographic and Health Survey (BDHS). Body mass index (BMI) has been used to measure the nutritional status of women. The kernel density was used to visualize the nutritional gap. The Oaxaca-Blinder (OB) decomposition method was used to ascertain influential determinants and their contributions to the existing gap between the richest-poorest classes of women.ResultsWe analyzed the data of 18,682 reproductive-aged women. There was a significant mean BMI gap of 4.1 unit (95% CI: 3.90-4.35) between the poorest-richest (25.6 vs 21.5) women. The overall prevalence of underweight, overweight and obese were 11.8%, 33.8% and 15.4%, respectively. The richest women were less underweight (7.5%) but more overweight (23.7%) and obese (42.2%). In contrast, the poorest women were more underweight (32.0%) but less overweight (13.9%) and obese (7.0%). According to results of OB decomposition method, all predictors combinedly can explain 1.62 units (95% CI: 1.31-1.93) of the total mean BMI gap (equivalent to 40%). Some of the major predictors were women years of education (0.45 units, 95% CI: 0.27-0.64), spouse years of education (0.16 units, 95% CI: -0.02-0.34), current working status (0.17 units, 95% CI: 0.10-0.34), access to Television (0.50 units, 95% CI: 0.28-0.72), and place of residence (0.37 units, 95% CI: 0.22-0.72). The unexplained part of the poorest-richest gap was 2.51 units (95% CI: 2.13-2.89), which means that this particular gap will remain unchanged even though the mean difference of the predictors was diminished.ConclusionsA large part of the nutritional gap (approximately 60%) between the poorest and richest classes of women are found to be unchanged by the predictors of the study. Therefore, further predictors should be identified to minimize such gap. Moreover, policy makers and relevant stakeholders should implement feasible strategies to minimize the existing differences in the major predictors.

Dataset Information

Marginal Contribution Feature Importance - an Axiomatic Approach for Explaining Data.

Publications

Marginal Contribution Feature Importance - an Axiomatic Approach for Explaining Data.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets