Massively Parallel Feature Extraction Framework Application in Predicting Dangerous Seismic Events

Abstract

Abstract-In this paper we introduce an automated mechanism for knowledge discovery from data streams. As a part of this work, we also present a new approach to the creation of classifiers ensemble based on a wide variety of models. Furthermore, we describe an innovative, highly scalable feature extraction and selection framework designed to work with the MapReduce programming model and the application of designed framework to build an ensemble of classifiers which takes into account both the quality and the diversity of individual models. The effectiveness of the solution has been verified through a participation in an open data mining competition which concerned the problem of predicting periods of increased seismic activity causing lifethreatening accidents in coal mines. The submitted solution obtained the highest AUC score of all the solutions uploaded by 106 participating research teams.

Predicting Dangerous Seismic Events: AAIA'16 Data Mining Challenge

“…In order to construct their solution, authors were using only the data available to all participants, however, due to their organizational involvement, team snm was excluded from the final ranking. More details regarding this solution can be found in [15].…”

Utilizing Frequent Pattern Mining for Solving Cold-Start Problem in Recommender Systems

“…Furthermore, we plan to consider changes in users' behavior and preferences by periodically updating frequent itemsets based on recent changes in rating history. It would also be of value to extend the users' and items' data representation by applying a more advanced feature extraction to model the similarities among them more effectively [44], [45], [46].…”

Considering various aspects of models’ quality in the ML pipeline - application in the logistics sector

“…One way to do this is to train models on diverse subsets of objects and attributes [17]. The training set selection, complemented by parallelization of computation, can lead to better general- ization and minimize the overall training latency [16], [18]. It is usually essential to ingest several diverse data sources to provide adequately rich data representation with many different attributes.…”

“…The first step in the developed pipeline is data ingestion and integration. Next, we perform data cleansing, encoding categorical variables to the numeric form, extracting some custom characteristics from text columns, imputing missing values, and conducting further feature extraction (FE) [18], [28].…”

“…Furthermore, the adopted principle of building a model on random subsets of attributes and rows allows to achieve a variety of different models within an ensemble [16], [17]. Such an approach to training set selection enables a very straightforward parallelization of the learning procedure and hence provides significant acceleration of computation [18].…”