Automated Feature Design for Numeric Sequence Classification by Genetic Programming

Abstract

Abstract-Pattern recognition methods rely on maximuminformation, minimum-dimension feature sets to reliably perform classification and regression tasks. Many methods exist to reduce feature set dimensionality and construct improved features from an initial set; however, there are few general approaches for design of features from numeric sequences. Any information lost in pre-processing or feature measurement cannot be recreated during pattern recognition. General approaches are needed to extend pattern recognition to include feature design and selection for numeric sequences, such as time series, within the learning process itself. This study proposes a novel genetic programming approach to automated feature design called Autofead. In this method, a genetic programming variant evolves a population of candidate features built from a library of sequence-handling functions. Numerical optimization methods, included through a hybrid approach, ensure that the fitness of candidate algorithms is measured using optimal parameter values. Autofead represents the first automated feature design system for numeric sequences to leverage the power and efficiency of both numerical optimization and standard pattern recognition algorithms. Potential applications include the monitoring of electrocardiogram signals for indications of heart failure, network traffic analysis for intrusion detection systems, vibration measurement for bearing condition determination in rotating machinery, and credit card activity for fraud detection.

Feature-Based Time-Series Analysis

“…A growing literature acknowledges that modern machine learning approaches can overcome some of the limitations of traditional data analysis, which is often plagued by subjective choices and small-scale comparison. This includes the use of large, interdisciplinary databases of features that can be compared systematically based on their empirical performance to automate feature selection, for example [4,35,51,58], and the use of ensemble methods that try to understand the properties of a time series or time-series dataset that make it suitable for a particular representation or algorithm [33,34,36,82]. These approaches acknowledge that no algorithm can perform well on all datasets [24,25], and use modern statistical approaches to tailor our methods to our data.…”

“…Comparative approaches to selecting global features for time series, described above, are limited to features that have already been developed and devised, i.e., there is no scope to devise completely new types of features for a given dataset. Automated feature construction for time series, such as the genetic programming (GP)-based approach Autofead (using combinations of interpretable transformations, like Fourier transforms, filtering, nonlinear transformations, and windowing) are powerful in their ability to adapt to particular data contexts to generate informative features [34]. However, features generated automatically in this way can be much more difficult to interpret, as they do not connect the data to interpretable areas of the timeseries analysis literature.…”

A Cost-Sensitive Deep Belief Network for Imbalanced Classification

“…In the second category, one can implement EA by optimizing classifiers for imbalanced classification at algorithmic level. Such approaches [45] optimize the classifier in the model space by using EA. Some studies [46], [47], [48] implement EAs to enhance rule-based classifiers.…”

Database Implementation and Testing of Dynamic Credit Card Fraud Detection System

“…etc. [2][3] [7] [10]. Véronique Van Vlasselaer and Cristián Bravo [5] has suggested the approach which combines inherent attributes derived from the characteristics of incoming transactions and the customer spending history using the primary characteristics such as RecencyFrequency-Monetary.…”