Subgroup Discovery Algorithms Research Articles

Comparison of Medical Opinions About the Decrease in Autopsies in Mexican Hospitals Using Data Mining

Subgroup discovery (SD) is a data mining technique that allows us to obtain the properties of each element given a particular population; these properties are of interest for a specific study, finding the most important or significant subgroups of the population. Also, the larger the population, the more successful the analysis and the creation of the subgroups, since, on this basis, the possibility of finding more unusual characteristics among the elements of the population is greater. The principal purpose of SD is not to obtain a predictive function, but to achieve a result that users can comprehend and interpret easily, and at the same time provide a more complete and suggestive description of the data. In this paper, we present an application of this technique to the medical field to analyze the opinions of physicians on the decreasing rates of autopsies in Mexican hospitals, utilizing five SD algorithms. The results obtained are the rules that allow for the comparison of medical opinions in three hospitals.

Using Subgroup Discovery to Relate Odor Pleasantness and Intensity to Peripheral Nervous System Reactions

Activation of the autonomic nervous system is a primary characteristic of human hedonic responses to sensory stimuli. For smells, general tendencies of physiological reactions have been described using classical statistics. However, these physiological variations are generally not quantified precisely; each psychophysiological parameter has very often been studied separately and individual variability was not systematically considered. The current study presents an innovative approach based on data mining, whose goal is to extract knowledge from a dataset. This approach uses a subgroup discovery algorithm which allows extraction of rules that apply to as many olfactory stimuli and individuals as possible. These rules are described by intervals on a set of physiological attributes. Results allowed both quantifying how each physiological parameter relates to odor pleasantness and perceived intensity but also describing the participation of each individual to these rules. This approach can be applied to other fields of affective sciences characterized by complex and heterogeneous datasets.

Automated identification of patient subgroups: A case-study on mortality of COVID-19 patients admitted to the ICU

Background– Subgroup discovery (SGD) is the automated splitting of the data into complex subgroups. Various SGD methods have been applied to the medical domain, but none have been extensively evaluated. We assess the numerical and clinical quality of SGD methods. Method– We applied the improved Subgroup Set Discovery (SSD++), Patient Rule Induction Method (PRIM) and APRIORI – Subgroup Discovery (APRIORI-SD) algorithms to obtain patient subgroups on observational data of 14,548 COVID-19 patients admitted to 73 Dutch intensive care units. Hospital mortality was the clinical outcome. Numerical significance of the subgroups was assessed with information-theoretic measures. Clinical significance of the subgroups was assessed by comparing variable importance on population and subgroup levels and by expert evaluation. Results– The tested algorithms varied widely in the total number of discovered subgroups (5-62), the number of selected variables, and the predictive value of the subgroups. Qualitative assessment showed that the found subgroups make clinical sense. SSD++ found most subgroups (n = 62), which added predictive value and generally showed high potential for clinical use. APRIORI-SD and PRIM found fewer subgroups (n = 5 and 6), which did not add predictive value and were clinically less relevant. Conclusion– Automated SGD methods find clinical subgroups that are relevant when assessed quantitatively (yield added predictive value) and qualitatively (intensivists consider the subgroups significant). Different methods yield different subgroups with varying degrees of predictive performance and clinical quality. External validation is needed to generalize the results to other populations and future research should explore which algorithm performs best in other settings.

Development of an algorithm-based approach using neuromuscular test results to indicate an increased risk for non-contact lower limb injuries in elite football players

ObjectivesThis retrospective cohort study explored an algorithm-based approach using neuromuscular test results to indicate an increased risk for non-contact lower limb injuries in elite football players.MethodsNeuromuscular data (eccentric hamstring strength,...

Identification of predictive factors of diabetic ketoacidosis in type 1 diabetes using a subgroup discovery algorithm.

To identify predictive factors for diabetic ketoacidosis (DKA) by retrospective analysis of registry data and the use of a subgroup discovery algorithm. Data from adults and children with type 1 diabetes and more than two diabetes-related visits were analysed from the Diabetes Prospective Follow-up Registry. Q-Finder, a supervised non-parametric proprietary subgroup discovery algorithm, was used to identify subgroups with clinical characteristics associated with increased DKA risk. DKA was defined as pH less than 7.3 during a hospitalization event. Data for 108 223 adults and children, of whom 5609 (5.2%) had DKA, were studied. Q-Finder analysis identified 11 profiles associated with an increased risk of DKA: low body mass index standard deviation score; DKA at diagnosis; age 6-10 years; age 11-15 years; an HbA1cof 8.87% or higher (≥ 73 mmol/mol); no fast-acting insulin intake; age younger than 15 years and not using a continuous glucose monitoring system; physician diagnosis of nephrotic kidney disease; severe hypoglycaemia; hypoglycaemic coma; and autoimmune thyroiditis. Risk of DKA increased with the number of risk profiles matching patients' characteristics. Q-Finder confirmed common risk profiles identified by conventional statistical methods and allowed the generation of new profiles that may help predict patients with type 1 diabetes who are at a greater risk of experiencing DKA.

Immune-Related Gene Signatures to Predict the Effectiveness of Chemoimmunotherapy in Triple-Negative Breast Cancer Using Exploratory Subgroup Discovery

Triple-negative breast cancer (TNBC) is an aggressive subtype of breast cancer with limited therapeutic options. Although immunotherapy has shown potential in TNBC patients, clinical studies have only demonstrated a modest response. Therefore, the exploration of immunotherapy in combination with chemotherapy is warranted. In this project we identified immune-related gene signatures for TNBC patients that may explain differences in patients' outcomes after anti-PD-L1+chemotherapy treatment. First, we ran the exploratory subgroup discovery algorithm on the TNBC dataset comprised of 422 patients across 24 studies. Secondly, we narrowed down the search to twelve homogenous subgroups based on tumor mutational burden (TMB, low or high), relapse status (disease-free or recurred), tumor cellularity (high, low and moderate), menopausal status (pre- or post) and tumor stage (I, II and III). For each subgroup we identified a union of the top 10% of genotypic patterns. Furthermore, we employed a multinomial regression model to predict significant genotypic patterns that would be linked to partial remission after anti-PD-L1+chemotherapy treatment. Finally, we uncovered distinct immune cell populations (T-cells, B-cells, Myeloid, NK-cells) for TNBC patients with various treatment outcomes. CD4-Tn-LEF1 and CD4-CXCL13 T-cells were linked to partial remission on anti-PD-L1+chemotherapy treatment. Our informatics pipeline may help to select better responders to chemoimmunotherapy, as well as pinpoint the underlying mechanisms of drug resistance in TNBC patients at single-cell resolution.

Subgroup discovery in structural equation models.

Structural equation modeling is one of the most popular statistical frameworks in the social and behavioral sciences. Often, detection of groups with distinct sets of parameters in structural equation models (SEM) are of key importance for applied researchers, for example, when investigating differential item functioning for a mental ability test or examining children with exceptional educational trajectories. In the present article, we present a new approach combining subgroup discovery-a well-established toolkit of supervised learning algorithms and techniques from the field of computer science-with structural equation models termed SubgroupSEM. We provide an overview and comparison of three approaches to modeling and detecting heterogeneous groups in structural equation models, namely, finite mixture models, SEM trees, and SubgroupSEM. We provide a step-by-step guide to applying subgroup discovery techniques for structural equation models, followed by a detailed and illustrated presentation of pruning strategies and four subgroup discovery algorithms. Finally, the SubgroupSEM approach will be illustrated on two real data examples, examining measurement invariance of a mental ability test and investigating interesting subgroups for the mediated relationship between predictors of educational outcomes and the trajectories of math competencies in 5th grade children. The illustrative examples are accompanied by examples of the R package subgroupsem, which is a viable implementation of our approach for applied researchers. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Identification of Immuno-Targeted Combination Therapies Using Explanatory Subgroup Discovery for Cancer Patients with EGFR Wild-Type Gene.

Simple SummaryCancer immunotherapy is a form of cancer treatment that uses a person’s own immune system to prevent, control, and eliminate cancer. However, immunotherapy alone may not be effective, especially in patients with limited treatment options. Immuno-targeted combination therapies have a potential to create synergetic effects with improved health outcomes. Therefore, there is a growing interest in searching for therapeutic combinations that could extend the benefits of immunotherapy. In this study, we designed a computational method that facilitated the identification of effective combination therapies for cancer patients with few treatment options. We determined several specific drug targets that substantially increased the odds of stable disease versus progressive disease for head and neck cancer, lung cancer, and melanoma. The identified treatment combinations were targets in several clinical trials. Moreover, our approach has the potential to improve the selection of patients for immuno-targeted combination therapies and lead to an overall improvement in health outcomes for cancer patients with limited treatment options.(1) Background: Phenotypic and genotypic heterogeneity are characteristic features of cancer patients. To tackle patients’ heterogeneity, immune checkpoint inhibitors (ICIs) represent some the most promising therapeutic approaches. However, approximately 50% of cancer patients that are eligible for treatment with ICIs do not respond well, especially patients with no targetable mutations. Over the years, multiple patient stratification techniques have been developed to identify homogenous patient subgroups, although matching a patient subgroup to a treatment option that can improve patients’ health outcomes remains a challenging task. (2) Methods: We extended our Subgroup Discovery algorithm to identify patient subpopulations that could potentially benefit from immuno-targeted combination therapies in four cancer types: head and neck squamous carcinoma (HNSC), lung adenocarcinoma (LUAD), lung squamous carcinoma (LUSC), and skin cutaneous melanoma (SKCM). We employed the proportional odds model to identify significant drug targets and the corresponding compounds that increased the likelihood of stable disease versus progressive disease in cancer patients with the EGFR wild-type (WT) gene. (3) Results: Our pipeline identified six significant drug targets and thirteen specific compounds for cancer patients with the EGFR WT gene. Three out of six drug targets—FCGR2B, IGF1R, and KIT—substantially increased the odds of having stable disease versus progressive disease. Progression-free survival (PFS) of more than 6 months was a common feature among the investigated subgroups. (4) Conclusions: Our approach could help to better select responders for immuno-targeted combination therapies and improve health outcomes for cancer patients with no targetable mutations.

Using knockoffs for controlled predictive biomarker identification.

One of the key challenges of personalized medicine is to identify which patients will respond positively to a given treatment. The area of subgroup identification focuses on this challenge, that is, identifying groups of patients that experience desirable characteristics, such as an enhanced treatment effect. A crucial first step towards the subgroup identification is to identify the baseline variables (eg, biomarkers) that influence the treatment effect, which are known as predictive variables. Many subgroup discovery algorithms return importance scores that capture the variables' predictive strength. However, a major limitation of these scores is that they do not answer the core question: "Which variables are actually predictive?" With our work we answer this question by using the knockoff framework, which is a general framework for controlling the false discovery rate when performing prognostic variable selection. In contrast, our work is the first that uses knockoffs for predictive variable selection. We introduce two novel knockoff filters: one parametric, building on variable importance scores derived from a penalized linear regression model, and one non-parametric, building on causal forest variable importance scores. We conduct extensive simulations to validate performance of the proposed methodology and we also apply the proposed methods to data from a randomized clinical trial.

DISGROU: an algorithm for discontinuous subgroup discovery.

In this paper, we focus on the problem of the search for subgroups in numerical data. This approach aims to identify the subsets of objects, called subgroups, which exhibit interesting characteristics compared to the average, according to a quality measure calculated on a target variable. In this article, we present DISGROU, a new approach that identifies subgroups whose attribute intervals may be discontinuous. Unlike the main algorithms in the field, the originality of our proposal lies in the way it breaks down the intervals of the attributes during the subgroup research phase. The basic assumption of our approach is that the range of attributes defining the groups can be disjoint to improve the quality of the identified subgroups. Indeed the traditional methods in the field perform the subgroup search process only over continuous intervals, which results in the identification of subgroups defined over wider intervals thus containing some irrelevant objects that degrade the quality function. In this way, another advantage of our approach is that it does not require a prior discretization of the attributes, since it works directly on the numerical attributes. The efficiency of our proposal is first demonstrated by comparing the results with two algorithms that are references in the field and then by applying to a case study.

Feature range analysis

We propose a feature range analysis algorithm whose aim is to derive features that explain the response variable better than the original features. Moreover, for binary classification problems, and for regression problems where positive and negative samples can be defined (e.g., using a threshold value of the numeric response variable), our aim is to derive features that explain, characterize and isolate the positive samples or subsets of positive samples that have the same root cause. Each derived feature represents a single or multi-dimensional subspace of the feature space, where each dimension is specified as a feature range pair for numeric features, and as a feature-level pair for categorical features. We call these derived features range features. Unlike most rule learning and subgroup discovery algorithms, the response variable can be numeric, and our algorithm does not require a discretization of the response. The algorithm has been applied successfully to real-life root-causing tasks in chip design, manufacturing, and validation, at Intel. Furthermore, we propose and experimentally evaluate a number of heuristics for usage of range features in building predictive models, demonstrating that prediction accuracy can be improved for the majority of real-life proprietary and open-source datasets used in the evaluation.

Q-Finder: An Algorithm for Credible Subgroup Discovery in Clinical Data Analysis - An Application to the International Diabetes Management Practice Study.

Addressing the heterogeneity of both the outcome of a disease and the treatment response to an intervention is a mandatory pathway for regulatory approval of medicines. In randomized clinical trials (RCTs), confirmatory subgroup analyses focus on the assessment of drugs in predefined subgroups, while exploratory ones allow a posteriori the identification of subsets of patients who respond differently. Within the latter area, subgroup discovery (SD) data mining approach is widely used—particularly in precision medicine—to evaluate treatment effect across different groups of patients from various data sources (be it from clinical trials or real-world data). However, both the limited consideration by standard SD algorithms of recommended criteria to define credible subgroups and the lack of statistical power of the findings after correcting for multiple testing hinder the generation of hypothesis and their acceptance by healthcare authorities and practitioners. In this paper, we present the Q-Finder algorithm that aims to generate statistically credible subgroups to answer clinical questions, such as finding drivers of natural disease progression or treatment response. It combines an exhaustive search with a cascade of filters based on metrics assessing key credibility criteria, including relative risk reduction assessment, adjustment on confounding factors, individual feature’s contribution to the subgroup’s effect, interaction tests for assessing between-subgroup treatment effect interactions and tests adjustment (multiple testing). This allows Q-Finder to directly target and assess subgroups on recommended credibility criteria. The top-k credible subgroups are then selected, while accounting for subgroups’ diversity and, possibly, clinical relevance. Those subgroups are tested on independent data to assess their consistency across databases, while preserving statistical power by limiting the number of tests. To illustrate this algorithm, we applied it on the database of the International Diabetes Management Practice Study (IDMPS) to better understand the drivers of improved glycemic control and rate of episodes of hypoglycemia in type 2 diabetics patients. We compared Q-Finder with state-of-the-art approaches from both Subgroup Identification and Knowledge Discovery in Databases literature. The results demonstrate its ability to identify and support a short list of highly credible and diverse data-driven subgroups for both prognostic and predictive tasks.

Anytime mining of sequential discriminative patterns in labeled sequences

It is extremely useful to exploit labeled datasets not only to learn models and perform predictive analytics but also to improve our understanding of a domain and its available targeted classes. The subgroup discovery task has been considered for more than two decades. It concerns the discovery of patterns covering sets of objects having interesting properties, e.g., they characterize or discriminate a given target class. Though many subgroup discovery algorithms have been proposed for both transactional and numerical data, discovering subgroups within labeled sequential data has been much less studied. First, we propose an anytime algorithm SeqScout that discovers interesting subgroups w.r.t. a chosen quality measure. This is a sampling algorithm that mines discriminant sequential patterns using a multi-armed bandit model. For a given budget, it finds a collection of local optima in the search space of descriptions and thus, subgroups. It requires a light configuration and is independent from the quality measure used for pattern scoring. We also introduce a second anytime algorithm MCTSExtent that pushes further the idea of a better trade-off between exploration and exploitation of a sampling strategy over the search space. To the best of our knowledge, this is the first time that the Monte Carlo Tree Search framework is exploited in a sequential data mining setting. We have conducted a thorough and comprehensive evaluation of our algorithms on several datasets to illustrate their added value, and we discuss their qualitative and quantitative results.

1624-P: Family History of Type 1 Diabetes (T1D) and Comorbidity Status in Individuals with T1D: A Computational Deep Exploratory Data Mining Approach

T1D is a progressive, autoimmune-mediated condition associated with numerous comorbidities and significant premature mortality. The offspring of fathers diagnosed with T1D exhibit a 2-3 times higher risk of developing T1D than the offspring of mothers diagnosed with T1D. Individuals with T1D who have an affected father present with a more aggressive disease phenotype at diagnosis. Yet the impact of paternal T1D (P-T1D) vs. maternal T1D (M-T1D) on health outcomes in individuals with T1D is unknown. We previously developed and validated a subgroup discovery algorithm that identifies highly contrasted patterns occurring with a significant difference in prevalence between two subgroups (e.g., P-T1D vs. M-T1D). Here we used this computational deep exploratory data mining method to analyze publicly-accessible data from the T1D Exchange Clinic Registry (2010-12; 2015-17). We used an Apache Spark high performance computing environment to apply our algorithm to family history, gender, and medical conditions data in the Registry. We identified multiple phenotypic contrasts between individuals with P-T1D (2010-12 data: n = 1011; 2015-17 data: n = 652) vs. M-T1D (2010-12 data: n = 528; 2015-17 data: n = 342). Using Fisher’s exact tests to determine statistical significance, P-T1D associated with hypothyroidism and male gender in 7.2% of cases (47/652), compared to 3.2% of cases (11/342) of M-T1D (p = 0.01). M-T1D was associated with hypertension (HTN) in 36.0% of cases (123/342), compared to 22.3% of cases (146/652) in P-T1D (p < 0.001). M-T1D was associated with diabetes-related neuropathy and HTN in 10.8% of cases (37/342), compared to 4.9% of cases (32/652) in P-T1D (p = 0.001). Complex patterns generated with this algorithm may yield contrast patterns that provide insights about personalized interventions to improve health outcomes. Whether contrast mining can be used to predict health trajectories of individuals with T1D remains to be determined. Disclosure E.M. Tallon: None. M.A. Clements: Consultant; Self; Glooko, Inc. Other Relationship; Self; Glooko, Inc. D. Liu: None. K. Boles: None. R.A. Stuck: None. C. Shyu: None. Funding National Institutes of Health (5T32LM012410)

Chemical features mining provides new descriptive structure-odor relationships

An important goal in researching the biology of olfaction is to link the perception of smells to the chemistry of odorants. In other words, why do some odorants smell like fruits and others like flowers? While the so-called stimulus-percept issue was resolved in the field of color vision some time ago, the relationship between the chemistry and psycho-biology of odors remains unclear up to the present day. Although a series of investigations have demonstrated that this relationship exists, the descriptive and explicative aspects of the proposed models that are currently in use require greater sophistication. One reason for this is that the algorithms of current models do not consistently consider the possibility that multiple chemical rules can describe a single quality despite the fact that this is the case in reality, whereby two very different molecules can evoke a similar odor. Moreover, the available datasets are often large and heterogeneous, thus rendering the generation of multiple rules without any use of a computational approach overly complex. We considered these two issues in the present paper. First, we built a new database containing 1689 odorants characterized by physicochemical properties and olfactory qualities. Second, we developed a computational method based on a subgroup discovery algorithm that discriminated perceptual qualities of smells on the basis of physicochemical properties. Third, we ran a series of experiments on 74 distinct olfactory qualities and showed that the generation and validation of rules linking chemistry to odor perception was possible. Taken together, our findings provide significant new insights into the relationship between stimulus and percept in olfaction. In addition, by automatically extracting new knowledge linking chemistry of odorants and psychology of smells, our results provide a new computational framework of analysis enabling scientists in the field to test original hypotheses using descriptive or predictive modeling.

Identification of Patient Profiles with Better Glycemic Outcomes Using Machine Learning—Results from the International Diabetes Management Practices Study (IDMPS)

Optimal diabetes care requires medical management, regular assessment, and self-care. We conducted a retrospective exploratory analysis using data from the IDMPS (a global observational survey on the management of diabetes in the developing world) using a subgroup discovery algorithm (Q-finder) to identify factors (patient profiles [PaPr]) associated with glycemic outcomes (good, HbA1c ≤7%; poor, HbA1c >7%) in people with type 2 diabetes. In total, 49,309 people from 48 countries in Latin America, Africa, the Middle East, Eurasia and Asia Pacific were included in seven individual waves of data collection. Variables analyzed (n=340) included demography, co-morbidities, clinical parameters and glucose lowering therapy; 107 were common to all waves. The algorithm identified combinations of variables and groups of people in which glycemic outcomes were significantly different to the total study population. Learning was conducted on Waves 6 and 7 (n=4937 and n=5695, respectively) to generate PaPr. PaPr were validated using a dataset that varied depending on the learning wave (Wave 6 max = 43,614; Wave 7 max = 44,372), and was dictated by the level of variable documentation in each wave. Each subgroup (PaPr) was characterized by 1-3 descriptive variables, and comprised ≥10% of the total patient population. All results reaching statistical significance (adjusted for multiple testing) were replicated in all waves. We have validated 11 out of 30 PaPr for clinical relevance. PaPr that positively affect glycemic outcomes include healthy diet and lifestyle, controlled FPG and PPG (calculated thresholds of 130 mg/dL and 180 mg/dL, respectively), systolic blood pressure <140 mmHg, absence of abnormal lipid profiles, and self-adjustment of insulin. Good glycemic control is often associated with management of cardiometabolic risk factors and self-management; a system-wide approach is needed to realize these goals. Disclosure J.C.N. Chan: Consultant; Self; Bayer AG. Other Relationship; Self; Bayer AG. Consultant; Self; Sanofi. Other Relationship; Self; Sanofi, Eli Lilly and Company, Amgen Inc.. Consultant; Self; AstraZeneca, Merck & Co., Inc., Pfizer Inc.. Other Relationship; Self; Pfizer Inc.. Board Member; Self; Asia Diabetes Foundation. Stock/Shareholder; Self; GemVCare. Other Relationship; Self; Merck Sharp & Dohme Corp.. Consultant; Self; Boehringer Ingelheim Pharmaceuticals, Inc., Novartis AG, Eli Lilly and Company. J.J. Gagliardino: Other Relationship; Self; Sanofi. H.M. Ilkova: Advisory Panel; Self; Abbott, Eli Lilly and Company, Novartis AG, Novo Nordisk A/S, Boehringer Ingelheim GmbH, Servier, Merck Sharp & Dohme Corp.. Speaker's Bureau; Self; AstraZeneca. Advisory Panel; Self; Medtronic, Roche Diabetes Care Health and Digital Solutions, Bilim İlac. F.J. Lavalle-Gonzalez: Advisory Panel; Self; Sanofi. Speaker's Bureau; Self; Sanofi. Research Support; Self; Janssen Pharmaceuticals, Inc.. Advisory Panel; Self; Janssen Pharmaceuticals, Inc.. Speaker's Bureau; Self; Janssen Pharmaceuticals, Inc.. Consultant; Self; Novo Nordisk Inc.. Advisory Panel; Self; AstraZeneca. Speaker's Bureau; Self; AstraZeneca. Research Support; Self; AstraZeneca. Advisory Panel; Self; Merck Sharp & Dohme Corp.. Research Support; Self; Boehringer Ingelheim Pharmaceuticals, Inc.. Advisory Panel; Self; Boehringer Ingelheim Pharmaceuticals, Inc.. Speaker's Bureau; Self; Boehringer Ingelheim Pharmaceuticals, Inc.. Advisory Panel; Self; Eli Lilly and Company. Speaker's Bureau; Self; Eli Lilly and Company. Advisory Panel; Self; Abbott. Speaker's Bureau; Self; Abbott. A. Ramachandran: None. G. Kaddaha: Other Relationship; Self; Sanofi. J. Mbanya: Advisory Panel; Self; GlaxoSmithKline plc.. Speaker's Bureau; Self; Novo Nordisk A/S, Sanofi, Servier. J. Chantelot: Employee; Self; Sanofi. Stock/Shareholder; Self; Sanofi. M.V. Shestakova: None. P. Aschner: Other Relationship; Self; Sanofi. F. Magnard: None. M. Rollot: None. A. Civet: None. F. Barbe: None.

Identifying key factors of student academic performance by subgroup discovery

Identifying the factors that influence student academic performance is essential to provide timely and effective support interventions. The data collected during enrolment and after commencement into a course provide an important source of information to assist with identifying potential risk indicators associated with poor academic performance and attrition. Both predictive and descriptive data mining techniques have been applied on educational data to discover the significant reasons behind student performance. These techniques have their own advantages and limitations. For example, predictive techniques tend to maximise accuracy for correctly classifying students, while the descriptive techniques simply search for interesting student features without considering their academic outcome. Subgroup discovery is a data mining method which takes the advantages of both predictive and descriptive approaches. This study uses subgroup discovery to extract significant factors of student performance for a certain outcome (Pass or Fail). In this work, we have utilised student demographic and academic data recorded at enrolment, as well as course assessment and participation data retrieved from the institution’s learning management system (Moodle) to detect the factors affecting student performance. The results have demonstrated the effectiveness of the subgroup discovery method in general in identifying the factors, and the pros and cons of some popular subgroup discovery algorithms used in this research. From the experiments, it has been found that students, who have indigent socio-economic background or been admitted based on special entry requirement, are most likely to fail. The experiments on Moodle data have revealed that students having lower level of access to the course resources and forum have higher possibility of being unsuccessful. From the combined data, we have identified some interesting subgroups which are not detected using enrolment or Moodle data separately. It has been found that those students, who study off-campus or part-time and have a low level of contributions to the course learning activities, are more likely to be the low-performing students.

Identifying consistent statements about numerical data with dispersion-corrected subgroup discovery

Existing algorithms for subgroup discovery with numerical targets do not optimize the error or target variable dispersion of the groups they find. This often leads to unreliable or inconsistent statements about the data, rendering practical applications, especially in scientific domains, futile. Therefore, we here extend the optimistic estimator framework for optimal subgroup discovery to a new class of objective functions: we show how tight estimators can be computed efficiently for all functions that are determined by subgroup size (non-decreasing dependence), the subgroup median value, and a dispersion measure around the median (non-increasing dependence). In the important special case when dispersion is measured using the mean absolute deviation from the median, this novel approach yields a linear time algorithm. Empirical evaluation on a wide range of datasets shows that, when used within branch-and-bound search, this approach is highly efficient and indeed discovers subgroups with much smaller errors.

Improvement of subgroup descriptions in noisy data by detecting exceptions

The presence of noise in datasets to which data mining techniques are applied can greatly reduce the quality and interest of the knowledge extracted. Subgroup discovery is a supervised descriptive rule discovery technique which is not exempt from this problem. The aim of this paper is to improve the descriptions of subgroups previously obtained by any subgroup discovery algorithm in noisy datasets. This is achieved using the post-processing approach of the MEFES algorithm, that first detects exceptions in the input subgroups and then includes those exceptions in the descriptions. The experiments performed in noisy datasets show the suitability of the proposal to improve the quality of the results.

MEFASD-BD: Multi-objective evolutionary fuzzy algorithm for subgroup discovery in big data environments - A MapReduce solution

Nowadays, there is an incredible increase of data volumes around the world, with the Internet as one of the main actors in this scenario and a growth rate above 30GB/s. The treatment of this huge amount of information cannot be carried out through traditional data mining algorithms in an efficient way and it is necessary to adapt and design new algorithms towards distributed paradigms such as MapReduce. This situation is a challenge for the community, investigated under the widely known term of big data.This paper presents a new algorithm for the subgroup discovery task called MEFASD-BD. The algorithm is developed in Apache Spark based on the MapReduce paradigm, and it is able to tackle high dimensional datasets in an efficient way. In fact, this algorithm is the first approximation to big data within evolutionary fuzzy systems for subgroup discovery. MEFASD-BD implements novel MapReduce functions which are able to analyse the quality of the subgroups obtained for each map with respect to the original dataset in order to improve the quality of these subgroups. In addition, the final reduce function of the algorithm employs the token competition operator in order to select the best rules extracted in the different maps. An experimental study with high dimensional datasets is performed in order to show the advantages of this algorithm in this type of problems. Specifically, the results of the study show an important reduction of the runtime while keeping the values in the standard quality measures for subgroup discovery.