A cluster analytic approach to identifying predictors and moderators of psychosocial treatment for bipolar depression: results from STEP-BD

Background We sought to address how predictors and moderators of psychotherapy for bipolar depression - identified individually in prior analyses - can inform the development of a metric for prospectively classifying treatment outcome in intensive psychotherapy (IP) versus collaborative care (CC) adjunctive to pharmacotherapy in the Systematic Treatment Enhancement Program (STEP-BD) study. Methods We conducted post-hoc analyses on 135 STEP-BD participants using cluster analysis to identify subsets of participants with similar clinical profiles and investigated this combined metric as a moderator and predictor of response to IP. We used agglomerative hierarchical cluster analyses and k-means clustering to determine the content of the clinical profiles. Logistic regression and Cox proportional hazard models were used to evaluate whether the resulting clusters predicted or moderated likelihood of recovery or time until recovery. Results The cluster analysis yielded a two-cluster solution: 1) "less-recurrent/severe" and 2) "chronic/recurrent." Rates of recovery in IP were similar for less-recurrent/severe and chronic/recurrent participants. Less-recurrent/severe patients were more likely than chronic/recurrent patients to achieve recovery in CC (p=.040, OR=4.56). IP yielded a faster recovery for chronic/recurrent participants, whereas CC led to recovery sooner in the less-recurrent/severe cluster (p=.034, OR=2.62). Limitations Cluster analyses require list-wise deletion of cases with missing data so we were unable to conduct analyses on all STEP-BD participants. Conclusions A well-powered, parametric approach can distinguish patients based on illness history and provide clinicians with symptom profiles of patients that confer differential prognosis in CC vs. IP.

A visual-numeric approach to clustering and anomaly detection for trajectory data

This paper proposes a novel application of Visual Assessment of Tendency (VAT)-based hierarchical clustering algorithms (VAT, iVAT, and clusiVAT) for trajectory analysis. We introduce a new clustering based anomaly detection framework named iVAT+ and clusiVAT+ and use it for trajectory anomaly detection. This approach is based on partitioning the VAT-generated Minimum Spanning Tree based on an efficient thresholding scheme. The trajectories are classified as normal or anomalous based on the number of paths in the clusters. On synthetic datasets with fixed and variable numbers of clusters and anomalies, we achieve 98 % classification accuracy. Our two-stage clusiVAT method is applied to 26,039 trajectories of vehicles and pedestrians from a parking lot scene from the real life MIT trajectories dataset. The first stage clusters the trajectories ignoring directionality. The second stage divides the clusters obtained from the first stage by considering trajectory direction. We show that our novel two-stage clusiVAT approach can produce natural and informative trajectory clusters on this real life dataset while finding representative anomalies.

A hybrid approach to clustering in big data

Clustering of big data has received much attention recently. In this paper, we present a new clusiVAT algorithm and compare it with four other popular data clustering algorithms. Three of the four comparison methods are based on the well known, classical batch k-means model. Specifically, we use k-means, single pass k-means, online k-means, and clustering using representatives (CURE) for numerical comparisons. clusiVAT is based on sampling the data, imaging the reordered distance matrix to estimate the number of clusters in the data visually, clustering the samples using a relative of single linkage (SL), and then noniteratively extending the labels to the rest of the data-set using the nearest prototype rule. Previous work has established that clusiVAT produces true SL clusters in compact-separated data. We have performed experiments to show that k-means and its modified algorithms suffer from initialization issues that cause many failures. On the other hand, clusiVAT needs no initialization, and almost always finds partitions that accurately match ground truth labels in labeled data. CURE also finds SL type partitions but is much slower than the other four algorithms. In our experiments, clusiVAT proves to be the fastest and most accurate of the five algorithms; e.g., it recovers 97% of the ground truth labels in the real world KDD-99 cup data (4 292 637 samples in 41 dimensions) in 76 s.

Dirichlet process mixture models with pairwise constraints for data clustering

The Dirichlet process mixture (DPM) model, a typical Bayesian nonparametric model, can infer the number of clusters automatically, and thus performing priority in data clustering. This paper investigates the influence of pairwise constraints in the DPM model. The pairwise constraint, known as two types: must-link (ML) and cannot-link (CL) constraints, indicates the relationship between two data points. We have proposed two relevant models which incorporate pairwise constraints: the constrained DPM (C-DPM) and the constrained DPM with selected constraints (SC-DPM). In C-DPM, the concept of chunklet is introduced. ML constraints are compiled into chunklets and CL constraints exist between chunklets. We derive the Gibbs sampling of the C-DPM based on chunklets. We further propose a principled approach to select the most useful constraints, which will be incorporated into the SC-DPM. We evaluate the proposed models based on three real datasets: 20 Newsgroups dataset, NUS-WIDE image dataset and Facebook comments datasets we collected by ourselves. Our SC-DPM performs priority in data clustering. In addition, our SC-DPM can be potentially used for short-text clustering.

A clustering-based failure mode and effect analysis model and its application to the edible bird nest industry

Failure mode and effect analysis (FMEA) is a popular safety and reliability analysis tool in examining potential failures of products, process, designs, or services, in a wide range of industries. While FMEA is a popular tool, the limitations of the traditional Risk Priority Number (RPN) model in FMEA have been highlighted in the literature. Even though many alternatives to the traditional RPN model have been proposed, there are not many investigations on the use of clustering techniques in FMEA. The main aim of this paper was to examine the use of a new Euclidean distance-based similarity measure and an incremental-learning clustering model, i.e., fuzzy adaptive resonance theory neural network, for similarity analysis and clustering of failure modes in FMEA; therefore, allowing the failure modes to be analyzed, visualized, and clustered. In this paper, the concept of a risk interval encompassing a group of failure modes is investigated. Besides that, a new approach to analyze risk ordering of different failure groups is introduced. These proposed methods are evaluated using a case study related to the edible bird nest industry in Sarawak, Malaysia. In short, the contributions of this paper are threefold: (1) a new Euclidean distance-based similarity measure, (2) a new risk interval measure for a group of failure modes, and (3) a new analysis of risk ordering of different failure groups.