Advances in Adaptive Data Analysis

An improved EEMD approach is introduced in this paper based on automatically obtaining the adding white noise amplitude and the ensemble number according to different analyzing signal characteristics. The adding white noise affects decomposition effect is researched in detail, a criterion of adding white noise in EEMD is established, and the improved EEMD algorithm is described. Simulated signals demonstrate the effectiveness of the improved EEMD in diagnosing the faults of rotating machinery. The improved EEMD is successfully applied to an early rub-impact fault detection of machine unit for catalytic cracking of heavy oil, with the fault reason being analyzed in detail. A gear fault detection of hot strip finishing mills is also analyzed utilizing the improved EEMD method. The results show that the improved EEMD can obtain more precise diagnosis results than the original EMD and FFT spectrum.

The estimation and significance testing of the first-order autoregressive (AR1) coefficient in short time series with trends are examined. The purpose is to identify the difficulties to which analysis procedures need to adjust for better results. The delta recursive AR1 estimator r_δ and the Sen–Theil trend estimator are viable for short sequence application. Significance testing for r_δ has low power. But the existence of trend has negligible influence in estimation and testing. The common practice of trend removal before AR1 estimation gives poorer results. Application to air quality data showed this could greatly change conclusions. Implication to analysis is discussed.

Discrete choice modeling (DCM) is widely used in economics, social studies, and marketing research for estimating utilities and preference probabilities of multiple alternatives. Data for the model is elicited from the respondents who are presented with several sets of items characterized by various attributes, and each respondent chooses the best alternative in each set. Estimation of utilities is usually performed in a multinomial-logit (MNL) modeling and software for Hierarchical Bayesian (HB) technique is usually applied to find individual utilities by iterative estimations. This paper describes an easy and convenient empirical Bayesian way to construct priors and combine them with the likelihood on individual level data. This allows the modeler to obtain posterior estimation of MNL utilities in noniterative evaluations. Logistic modeling for the posterior frequencies is performed using the linear link of their logarithm of odds that clarifies the results of DCM modeling. The problem of overfitting is considered and the optimum balance between signal and noise in the precision of individual prediction and the smoothing of overall data is suggested. Actual market research data are used and the results are discussed.

Sensors placed in the field frequently suffer faults or communication errors, causing readings to become unavailable for prediction purposes. Statistical approaches are commonly used to generate an artificial value to approximate a missing sensor reading and complete a feature vector for prediction. In this paper, we present a new multiple classifier system that is capable of predicting without artificial approximations of missing values. Each classifier in the system is trained on different feature subsets, so that a classifier exists for each potential combination of available features. We have evaluated the proposed approach on a set of benchmark datasets to establish the generic nature of the problem, and on a real-world environmental event detection application that inspires the need for this work. On benchmark datasets, we artificially vary the percentage of missing values (i) during the prediction phase, and (ii) during both the training and prediction phases. We compare the performance of the proposed multiple classifier system against statistical approaches and provide a guideline on their suitability under the above scenarios.

A particular quintet singular valued decomposition (Quintet-SVD) is introduced in this paper via empirical mode decompositions (EMDs). The Quintet-SVD results in four specific orthogonal matrices with a diagonal matrix of singular values. Furthermore, this paper shows relationships between the Quintet-SVD and traditional SVD, generalized low rank approximations of matrices (GLRAM) of one single matrix, and EMDs. One application of the Quintet-SVD for speech enhancement is shown and compared with an application of traditional SVD.