Matt Triff

Identifying users and activities with cognitive signal processing from a wearable headband

This paper studies the supervised classification of electroencephalogram (EEG) brain signals to identify persons and their activities. The brain signals are obtained from a commercially available and modestly priced wearable headband. Such wearable devices generate a large amount of data and due to their attractive pricing structure are becoming increasingly commonplace. As a result, the data generated from such wearables will increase exponentially leading to many interesting data mining opportunities. We propose a representation that reduces variable length signals to a more manageable and uniformly fixed length distributions. These fixed length distributions can then be used with a variety of data mining techniques. The experiments with a number of classification techniques, including decision trees, SVM, neural networks, and random forests show that it is possible to identify both the persons and the activities with a reasonable degree of precision.

Advances in Rough and Soft Clustering: Meta-Clustering, Dynamic Clustering, Data-Stream Clustering

Over the last five decades, clustering has established itself as a primary unsupervised learning technique. In most major data mining projects clustering can serve as a first step in understanding the available data. Clustering is used for creating meaningful profiles of entities in an application. It can also be used to compress the dataset into more manageable granules. The initial methods of crisp clustering objects represented using numeric attributes have evolved to address the demands of the real-world. These extensions include the use of soft computing techniques such as fuzzy and rough set theory, the use of centroids and medoids for computational efficiency, modes to accommodate categorical attributes, dynamic and stream clustering for managing continuous accumulation of data, and meta-clustering for correlating parallel clustering processes. This paper uses applications in engineering, web usage, retail, finance, and social networks to illustrate some of the recent advances in clustering and their role in improved profiling, as well as augmenting prediction, classification, association mining, dimensionality reduction, and optimization tasks.

Clustering Ensemble for Prioritized Sampling Based on Average and Rough Patterns

This paper proposes a clustering ensemble for prioritized sampling to tackle a big data problem. The proposal first creates separate clustering schemes of objects using different dimensions of the dataset. These clustering schemes are then combined to create a representative sample based on all the possible combinations of profiles. The resulting clustering ensemble will help system developers to reduce the number of objects that need to be analyzed while making sure that all the profile combinations are comprehensively covered. The proposal further ranks the objects in the sample based on their ability to capture important aspects of each of the criteria. The proposed approach can be used to provide a priority based analysis/modelling over an extended period of time. The prioritized analysis/models will be available for use in a reasonably short period of time. The quality of the analysis/modelling will continuously improve as more and more objects in the sample are processed according to their rank in the sample. The proposal is applied to a large set of weather stations to create a ranked sample based on hourly and monthly variations of important weather parameters, such as temperature, solar radiation, wind speed, and humidity. The experiments also demonstrate how a combination of average and rough patterns help in creating more meaningful profiles.

Evolutionary semi-supervised rough categorization of brain signals from a wearable headband

This paper explores the possibility of using distance based semi-supervised learning for creating lower and upper approximations of biometric signals. An evolutionary approach applied to both crisp and rough clustering optimizes both the within cluster scatter and the precision of the classification. The proposed approach is demonstrated through data collected from a wearable headband that recorded EEG brain signals. The brain signals are recorded for a number of participants performing various tasks. The approach identifies medoids that can best identify the participants. The evolutionary semi-supervised crisp and rough clustering is shown to favorably compare with the conventional unsupervised algorithms such as K-means.

Fuzzy and Crisp Recursive Profiling of Online Reviewers and Businesses

Users of online review sites can benefit from knowing the profiles of the businesses, as well as the profiles of reviewers who reviewed the businesses. This paper describes crisp and fuzzy metaclustering techniques to evolve two recursively defined clustering schemes of both businesses and reviewers in parallel using a real-world dataset supplied by yelp.com. The objective is to profile the businesses and reviewers by grouping them based on similar characteristics. The novelty of the proposed approach is in the fact that the representations of both businesses and reviewers change dynamically throughout the metaclustering process. A business is represented by static information obtained from the database and dynamic information obtained from the clustering of reviewers who reviewed the business. Similarly, the reviewer representation augments the static representation from the database with profiles of businesses who are reviewed by these reviewers. The resulting web-based service provides a facility for users to find similar businesses/reviewers based on the category of the business, rating, number of reviews, and number of check-ins. It also provides a succinct profile of a business or reviewer based on these factors so that the users can put the reviews in context. Since an object can belong to multiple clusters in fuzzy metaclustering, it is possible to absorb some of the extreme groups consisting of outliers in one of the mainstream clusters. As a result, the fuzzy metaclustering leads to more uniformly distributed and moderate profiles.

Fuzzy Clustering Ensemble for Prioritized Sampling Based on Average and Rough Patterns

This paper uses fuzzy clustering to extend a previous prioritized sampling proposal. In many big data problems, modeling an individual object such as a large engineering plant can be a tedious process requiring up to a month of analysis. A solution is to model as many representative objects as possible to represent the entire population. A new object can then use a model (or combination of models) from previously analyzed objects that best matches its characteristics. Since the modeling process can continue indefinitely adding models over time, we prioritize the sampling based on the ability of objects to represent as many characteristics as possible. The approach is demonstrated with a large set of weather stations to create a ranked sample based on hourly and monthly variations of important weather parameters, such as temperature, solar radiation, wind speed, and humidity. The weather patterns are represented using a combination of average and rough patterns to capture the essence of the distribution. The weather stations are grouped using Fuzzy C-Means and the objects with the largest fuzzy memberships are used as the representatives of each cluster. The weather stations representing a combination of different clustering schemes are then ranked based on the number of weather patterns they represent.

Nonlinear classification, linear clustering, evolutionary semi-supervised three-way decisions: A comparison

This paper compares the semantically meaningful machine learning algorithms with the black box models. The machine learning models are applied to a real world wearable dataset for biometric identification of individuals. The semantically meaningful decision tree is compared with more accurate black-box models such as neural networks, random forest, and support vector machines. The paper further explores the possibility of using unsupervised learning that uses linear distances for separating the categories. Since the distance from the center is used to delineate the clusters, the centroids of the unsupervised clusters provide a semantic profile of the categories. The crisp K-means clustering is enhanced with evolutionary algorithms that primarily uses the distance from the center as the primary criteria, but nudges the clustering towards known classification using a semi-supervised penalty. Finally, the use of rough sets is shown to provide notable semantic information with the help of the three-way decision principle.

Recursive Profiles of Businesses and Reviewers on Yelp.com

This paper uses a novel recursive meta-profiling technique where profiles from one set of objects dynamically change the representation of another set of objects. Two profiling schemes evolve in parallel influencing each other through indirect recursion. This is demonstrated with the help of a yelp.com dataset consisting of businesses and reviewers. A business is represented by static information obtained from the database and dynamic information obtained from clustering of reviewers who reviewed the business. Similarly, the reviewer representation augments the static representation from the database with profiles of businesses who are reviewed by these reviewers. The resulting service provides a facility for users to find similar businesses/reviewers based on the grading of the business, easy/hard grading, and types of businesses. It also provides a succinct profile of business/reviewer based on these factors, so users can put the reviews in context.