An online outlier identification and removal scheme for improving fault detection performance

Development and validation of an equation to predict the incidence of coronary heart disease in patients with type 2 diabetes in Japan

Objective

In the diabetes treatment policy after the Kumamoto Declaration 2013, it is difficult to accurately predict the incidence of complications in patients using the JJ risk engine. This study was conducted to develop a prediction equation suitable for the current diabetes treatment policy using patient data from Kitasato University Kitasato Institute Hospital (Hospital A) and to externally validate the developed equation using patient data from Kitasato University Hospital (Hospital B). Outlier tests were performed on the patient data from Hospital A to exclude the outliers. Prediction equation was developed using the patient data excluding the outliers and was subjected to external validation.

Results

By excluding outlier data, we could develop a new prediction equation for the incidence of coronary heart disease (CHD) as a complication of type 2 diabetes, incorporating the use of antidiabetic drugs with a high risk of hypoglycemia. This is the first prediction equation in Japan that incorporates the use of antidiabetic drugs. We believe that it will be useful in preventive medicine for treatment for people at high risk of CHD as a complication of diabetes or other diseases. In the future, we would like to confirm the accuracy of this equation at other facilities.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13104-021-05844-w.

Online Anomaly Detection With Bandwidth Optimized Hierarchical Kernel Density Estimators

We propose a novel unsupervised anomaly detection algorithm that can work for sequential data from any complex distribution in a truly online framework with mathematically proven strong performance guarantees. First, a partitioning tree is constructed to generate a doubly exponentially large hierarchical class of observation space partitions, and every partition region trains an online kernel density estimator (KDE) with its own unique dynamical bandwidth. At each time, the proposed algorithm optimally combines the class estimators to sequentially produce the final density estimation. We mathematically prove that the proposed algorithm learns the optimal partition with kernel bandwidths that are optimized in both region-specific and time-varying manner. The estimated density is then compared with a data-adaptive threshold to detect anomalies. Overall, the computational complexity is only linear in both the tree depth and data length. In our experiments, we observe significant improvements in anomaly detection accuracy compared with the state-of-the-art techniques.

Efficient treatment of outliers and class imbalance for diabetes prediction

Learning from outliers and imbalanced data remains one of the major difficulties for machine learning classifiers. Among the numerous techniques dedicated to tackle this problem, data preprocessing solutions are known to be efficient and easy to implement. In this paper, we propose a selective data preprocessing approach that embeds knowledge of the outlier instances into artificially generated subset to achieve an even distribution. The Synthetic Minority Oversampling TEchnique (SMOTE) was used to balance the training data by introducing artificial minority instances. However, this was not before the outliers were identified and oversampled (irrespective of class). The aim is to balance the training dataset while controlling the effect of outliers. The experiments prove that such selective oversampling empowers SMOTE, ultimately leading to improved classification performance.

Probabilistic Distance for Mixtures of Independent Component Analyzers

Independent component analysis (ICA) is a blind source separation technique where data are modeled as linear combinations of several independent non-Gaussian sources. The independence and linear restrictions are relaxed using several ICA mixture models (ICAMMs) obtaining a two-layer artificial neural network structure. This allows for dependence between sources of different classes, and thus, a myriad of multidimensional probability density functions can be accurate modeled. This paper proposes a new probabilistic distance (PDI) between the parameters learned for two ICAMMs. The PDI is computed explicitly, unlike the popular Kullback-Leibler divergence (KLD) and other similar metrics, removing the need for numerical integration. Furthermore, the PDI is symmetric and bounded within 0 and 1, which enables its use as a posterior probability in fusion approaches. In this paper, the PDI is employed for change detection by measuring the distance between two ICAMMs learned in consecutive time windows. The changes might be associated with relevant states from a process under analysis that are explicitly reflected in the learned ICAMM parameters. The proposed distance was tested in two challenging applications using simulated and real data: 1) detecting flaws in materials using ultrasounds and 2) detecting changes in electroencephalography signals from humans performing neuropsychological tests. The results demonstrate that the PDI outperforms the KLD in change-detection capabilities.

A multiple kernel classification approach based on a Quadratic Successive Geometric Segmentation methodology with a fault diagnosis case

This work presents a new approach for solving classification and learning problems. The Successive Geometric Segmentation technique is applied to encapsulate large datasets by using a series of Oriented Bounding Hyper Box (OBHBs). Each OBHB is obtained through linear separation analysis and each one represents a specific region in a pattern's solution space. Also, each OBHB can be seen as a data abstraction layer and be considered as an individual Kernel. Thus, it is possible by applying a quadratic discriminant function, to assemble a set of nonlinear surfaces separating each desirable pattern. This approach allows working with large datasets using high speed linear analysis tools and yet providing a very accurate non-linear classifier as final result. The methodology was tested using the UCI Machine Learning repository and a Power Transformer Fault Diagnosis real scenario problem. The results were compared with different approaches provided by literature and, finally, the potential and further applications of the methodology were also discussed.

An Integrated Learning and Filtering Approach for Fault Diagnosis of a Class of Nonlinear Dynamical Systems

This paper develops an integrated filtering and adaptive approximation-based approach for fault diagnosis of process and sensor faults in a class of continuous-time nonlinear systems with modeling uncertainties and measurement noise. The proposed approach integrates learning with filtering techniques to derive tight detection thresholds, which is accomplished in two ways: 1) by learning the modeling uncertainty through adaptive approximation methods and 2) by using filtering for dampening measurement noise. Upon the detection of a fault, two estimation models, one for process and the other for sensor faults, are initiated in order to identify the type of fault. Each estimation model utilizes learning to estimate the potential fault that has occurred, and adaptive isolation thresholds for each estimation model are designed. The fault type is deduced based on an exclusion-based logic, and fault detectability and identification conditions are rigorously derived, characterizing quantitatively the class of faults that can be detected and identified by the proposed scheme. Finally, simulation results are used to demonstrate the effectiveness of the proposed approach.

A one-class kernel fisher criterion for outlier detection

Recently, Dufrenois and Noyer proposed a one class Fisher's linear discriminant to isolate normal data from outliers. In this paper, a kernelized version of their criterion is presented. Originally on the basis of an iterative optimization process, alternating between subspace selection and clustering, I show here that their criterion has an upper bound making these two problems independent. In particular, the estimation of the label vector is formulated as an unconstrained binary linear problem (UBLP) which can be solved using an iterative perturbation method. Once the label vector is estimated, an optimal projection subspace is obtained by solving a generalized eigenvalue problem. Like many other kernel methods, the performance of the proposed approach depends on the choice of the kernel. Constructed with a Gaussian kernel, I show that the proposed contrast measure is an efficient indicator for selecting an optimal kernel width. This property simplifies the model selection problem which is typically solved by costly (generalized) cross-validation procedures. Initialization, convergence analysis, and computational complexity are also discussed. Lastly, the proposed algorithm is compared with recent novelty detectors on synthetic and real data sets.