Gladys Castillo

Learning with Drift Detection

Most of the work in machine learning assume that examples are generated at random according to some stationary probability distribution. In this work we study the problem of learning when the distribution that generate the examples changes over time. We present a method for detection of changes in the probability distribution of examples. The idea behind the drift detection method is to control the online error-rate of the algorithm. The training examples are presented in sequence. When a new training example is available, it is classified using the actual model. Statistical theory guarantees that while the distribution is stationary, the error will decrease. When the distribution changes, the error will increase. The method controls the trace of the online error of the algorithm. For the actual context we define a warning level, and a drift level. A new context is declared, if in a sequence of examples, the error increases reaching the warning level at example k w , and the drift level at example k d . This is an indication of a change in the distribution of the examples. The algorithm learns a new model using only the examples since k w . The method was tested with a set of eight artificial datasets and a real world dataset. We used three learning algorithms: a perceptron, a neural network and a decision tree. The experimental results show a good performance detecting drift and with learning the new concept. We also observe that the method is independent of the learning algorithm.

Learning with local drift detection

Most of the work in Machine Learning assume that examples are generated at random according to some stationary probability distribution. In this work we study the problem of learning when the distribution that generates the examples changes over time. We present a method for detection of changes in the probability distribution of examples. The idea behind the drift detection method is to monitor the online error-rate of a learning algorithm looking for significant deviations. The method can be used as a wrapper over any learning algorithm. In most problems, a change affects only some regions of the instance space, not the instance space as a whole. In decision models that fit different functions to regions of the instance space, like Decision Trees and Rule Learners, the method can be used to monitor the error in regions of the instance space, with advantages of fast model adaptation. In this work we present experiments using the method as a wrapper over a decision tree and a linear model, and in each internal-node of a decision tree. The experimental results obtained in controlled experiments using artificial data and a real-world problem show a good performance detecting drift and in adapting the decision model to the new concept.

A Methodology for Developing Adaptive Educational-Game Environments

In this paper we present a methodology for describing adaptive educational-game environments and a model that supports the environment design process. These environments combine the advantages of educational games with those derived from the adaptation. The proposed methodology allows the specification of educational methods that can be used for the game environment generation. The educational goals, the activities that the users can perform, their organization and sequencing, along with the games to be played and the game stories are selected or dynamically generated taking into account the users features and behaviors.

Designing a Dynamic Bayesian Network for Modeling Students' Learning Styles

When using learning object repositories, it is interesting to have mechanisms to select the more adequate objects for each student. For this kind of adaptation, it is important to have sound models to estimate the relevant features. In this paper we present a student model to account for learning styles, based on the model defined by Felder and Sylverman and implemented using dynamic Bayesian networks. The model is initialized according to the results obtained by the student in the index of learning styles questionnaire, and then fine-tuned during the course of the interaction using the Bayesian model, The model is then used to classify objects in the repository as appropriate or not for a particular student.

Machine Learning algorithms applied to the classification of robotic soccer formations and opponent teams

Machine Learning (ML) and Knowledge Discovery (KD) are research areas with several different applications but that share a common objective of acquiring more and new information from data. This paper presents an application of several ML techniques in the identification of the opponent team and also on the classification of robotic soccer formations in the context of RoboCup international robotic soccer competition. RoboCup international project includes several distinct leagues were teams composed by different types of real or simulated robots play soccer games following a set of pre-established rules. The simulated 2D league uses simulated robots encouraging research on artificial intelligence methodologies like high-level coordination and machine learning techniques. The experimental tests performed, using four distinct datasets, enabled us to conclude that the Support Vector Machines (SVM) technique has higher accuracy than the k-Nearest Neighbor, Neural Networks and Kernel Naïve Bayes in terms of adaptation to a new kind of data. Also, the experimental results enable to conclude that using the Principal Component Analysis SVM achieves worse results than using simpler methods that have as primary assumption the distance between samples, like k-NN.

An adaptive prequential learning framework for bayesian network classifiers

We introduce an adaptive prequential learning framework for Bayesian Network Classifiers which attempts to handle the cost-performance trade-off and cope with concept drift. Our strategy for incorporating new data is based on bias management and gradual adaptation. Starting with the simple Naive Bayes, we scale up the complexity by gradually increasing the maximum number of allowable attribute dependencies, and then by searching for new dependences in the extended search space. Since updating the structure is a costly task, we use new data to primarily adapt the parameters and only if this is really necessary, do we adapt the structure. The method for handling concept drift is based on the Shewhart P-Chart. We evaluated our adaptive algorithms on artificial domains and benchmark problems and show its advantages and future applicability in real-world on-line learning systems.

Adaptive bayes for a student modeling prediction task based on learning styles

We present Adaptive Bayes, an adaptive incremental version of Naive Bayes, to model a prediction task based on learning styles in the context of an Adaptive Hypermedia Educational System. Since the students preferences can change over time, this task is related to a problem known as concept drift in the machine learning community. For this class of problems an adaptive predictive model, able to adapt quickly to the users changes, is desirable. The results from conducted experiments show that Adaptive Bayes seems to be a fine and simple choice for this kind of prediction task in user modeling.

Adaptation to drifting concepts

Most of supervised learning algorithms assume the stability of the target concept over time. Nevertheless in many real-user modeling systems, where the data is collected over an extended period of time, the learning task can be complicated by changes in the distribution underlying the data. This problem is known in machine learning as concept drift. The main idea behind Statistical Quality Control is to monitor the stability of one or more quality characteristics in a production process which generally shows some variation over time. In this paper we present a method for handling concept drift based on Shewhart P-Charts in an on-line framework for supervised learning. We explore the use of two alternatives P-charts, which differ only by the way they estimate the target value to set the center line. Experiments with simulated concept drift scenarios in the context of a user modeling prediction task compare the proposed method with other adaptive approaches. The results show that, both P-Charts consistently recognize concept changes, and that the learner can adapt quickly to these changes to maintain its performance level.

TF-SIDF: Term frequency, sketched inverse document frequency

Exact calculation of the TF-IDF weighting function in massive streams of documents involves challenging memory space requirements. In this work, we propose TF-SIDF, a novel solution for extracting relevant words from streams of documents with a high number of terms. TF-SIDF relies on the Count-Min Sketch data structure, which allows to estimate the counts of all the terms in the stream. Results of the experiments conducted with two dataset show that this sketch-based algorithm achieves good approximations of the TF-IDF weighting values (as a rule, the top terms with highest TF-IDF values remaining the same), while substantial savings in memory usage are observed. It is also observed that the performance is highly correlated with the sketch size, and that wider sketch configurations are preferable given the same sketch size.

Bias management of bayesian network classifiers

The purpose of this paper is to describe an adaptive algorithm for improving the performance of Bayesian Network Classifiers (BNCs) in an on-line learning framework. Instead of choosing a priori a particular model class of BNCs, our adaptive algorithm scales up the models complexity by gradually increasing the number of allowable dependencies among features. Starting with the simple Naive Bayes structure, it uses simple decision rules based on qualitative information about the performances dynamics to decide when it makes sense to do the next move in the spectrum of feature dependencies and to start searching for a more complex classifier. Results in conducted experiments using the class of Dependence Bayesian Classifiers on three large datasets show that our algorithm is able to select a model with the appropriate complexity for the current amount of training data, thus balancing the computational cost of updating a model with the benefits of increasing in accuracy.