Carlos J. Alonso

Interval and dynamic time warping-based decision trees

This work presents decision trees adequate for the classification of series data. There are several methods for this task, but most of them focus on accuracy. One of the requirements of data mining is to produce comprehensible models. Decision trees are one of the most comprehensible classifiers. The use of these methods directly on this kind of data is, generally, not adequate, because complex and inaccurate classifiers are obtained. Hence, instead of using the raw features, new ones are constructed.This work presents two types of trees. In interval-based trees, the decision nodes evaluate a function (e.g., the average) in an interval and the result is compared to a threshold. For DTW-based trees each decision node has a reference example. The distance from the example to classify to the reference example is calculated and then it is compared to a threshold.The method for obtaining these trees it is based on 1) to develop a method that obtains for a 2-class data set a classifier formed by a new feature (a function in an interval or the distance to a reference example) and a threshold, 2) to use the boosting method to obtain an ensemble of these classifiers, and 3) to use a method for constructing decision trees using as data set the features selected by boosting.

Support vector machines of interval-based features for time series classification

In previous works, a time series classification system has been presented. It is based on boosting very simple classifiers, formed only by one literal. The used literals are based on temporal intervals. The obtained classifiers were simply a linear combination of literals, so it is natural to expect some improvements in the results if those literals were combined in more complex ways. In this work we explore the possibility of using the literals selected by the boosting algorithm as new features, and then using a SVM with these metafeatures. The experimental results show the validity of the proposed method.

Enhancing Consistency Based Diagnosis with Machine Learning Techniques

This paper proposes a diagnosis architecture that integrates consistency based diagnosis with induced time series classifiers, trying to combine the advantages of both methods. Consistency based diagnosis allows fault detection and localization without prior knowledge of the device fault modes. Machine learning techniques are able to induce time series classifiers that may be used to identify fault modes of a dynamic systems. The diagnostician performs fault detection and localization resorting to consistency based diagnosis through possible conflicts. Then, a time series classifier, induced from simulated examples, generates a sequence of faults modes, coherent with the result of the fault localization stage, and ordered by fault modes confidence. Finally, to simplify the diagnosis task, it is considered as a subtask of a supervisory system, who is in charge of identifying the working conditions for the physical system.

Bias and variance of rotation-based ensembles

In Machine Learning, ensembles are combination of classifiers. Their objective is to improve the accuracy. In previous works, we have presented a method for the generation of ensembles, named rotation-based. It transforms the training data set; it groups, randomly, the attributes in different subgroups, and applies, for each group, an axis rotation. If the used method for the induction of the classifiers is not invariant to rotations in the data set, the generated classifiers can be very different. In this way, different classifiers can be obtained (and combined) using the same induction method. The bias-variance decomposition of the error is used to get some insight into the behaviour of a classifier. It has been used to explain the success of ensemble learning techniques. In this work the bias and variance for the presented and other ensemble methods are calculated and used for comparison purposes.

Rotation-Based Ensembles

A new method for ensemble generation is presented. It is based on grouping the attributes in different subgroups, and to apply, for each group, an axis rotation, using Principal Component Analysis. If the used method for the induction of the classifiers is not invariant to rotations in the data set, the generated classifier can be very different. Hence, once of the objectives aimed when generating ensembles is achieved, that the different classifiers were rather diverse. The majority of ensemble methods eliminate some information (e.g., instances or attributes) of the data set for obtaining this diversity. The proposed ensemble method transforms the data set in a way such as all the information is preserved. The experimental validation, using decision trees as base classifiers, is favorable to rotation based ensembles when comparing to Bagging, Random Forests and the most well-known version of Boosting.

Diagnosis of dynamic systems: a knowledge model that allows tracking the system during the diagnosis process

A knowledge-based model for on-line diagnosis of complex dynamic systems is proposed. Domain knowledge is modelled via causal networks which consider temporal relationships among symptoms and causes. Inference and task knowledge is described using the CommonKADS methodology. The main feature of the proposal is that the diagnosis task is able to track the evolution of the system incorporating new symptoms to the diagnosis process. Diagnosis is conceived as a task to be carried out by a supervisory system, which could select the suitable causal network to perform diagnosis, depending on the current system configuration and operation point.

Stacking Dynamic Time Warping for the Diagnosis of Dynamic Systems

This paper explores an integrated approach to diagnosis of complex dynamic systems. Consistency-based diagnosis is capable of performing automatic fault detection and localization using just correct behaviour models. Nevertheless, it may exhibit low discriminative power among fault candidates. Hence, we combined the consistency based approach with machine learning techniques specially developed for fault identification of dynamic systems. In this work, we apply Stacking to generate time series classifiers from classifiers of its univariate time series components. The Stacking scheme proposed uses K-NN with Dynamic Time Warping as a dissimilarity measure for the level 0 learners and Naive Bayes at level 1. The method has been tested in a fault identification problem for a laboratory scale continuous process plant. Experimental results show that, for the available data set, the former Stacking configuration is quite competitive, compare to other methods like tree induction, Support Vector Machines or even K-NN and Naive Bayes as stand alone methods.

A Proposal of Diagnosis for an ITS for Computational Logic

We describe in this paper an ITS called SIAL that supports the learning of problem solving skills in computational logic from obtaining the clause form of simple well formed formulae to hyperresolution. The core function in SIAL is the error diagnosis module, that has the role of detecting and interpreting the mistakes of the learner while he/she is solving the exercises. It combines both model-based and knowledge-based (expertise) diagnosis in order to achieve more accurate results. SIAL complements this core function with a flexible user interface and a pedagogical module that offers three modes of interaction adapted to the learner’s level of expertise. SIAL is currently being tested by a group of volunteers in order to measure and tune its accuracy, as a preliminary step before performing tests in real conditions.

A representation of temporal aspects in knowledge based systems modelling: a monitoring example

One of the difficulties of using Artificial Neural Networks (ANNs) to estimate atmospheric temperature is the large number of potential input variables available. In this study, four different feature extraction methods were used to reduce the input vector to train four networks to estimate temperature at different atmospheric levels. The four techniques used were: genetic algorithms (GA), coefficient of determination (CoD), mutual information (MI) and simple neural analysis (SNA). The results demonstrate that of the four methods used for this data set, mutual information and simple neural analysis can generate networks that have a smaller input parameter set, while still maintaining a high degree of accuracy.