Xiuju Fu

Data dimensionality reduction with application to simplifying RBF network structure and improving classification performance

For high dimensional data, if no preprocessing is carried out before inputting patterns to classifiers, the computation required may be too heavy. For example, the number of hidden units of a radial basis function (RBF) neural network can be too large. This is not suitable for some practical applications due to speed and memory constraints. In many cases, some attributes are not relevant to concepts in the data at all. In this paper, we propose a novel separability-correlation measure (SCM) to rank the importance of attributes. According to the attribute ranking results, different attribute subsets are used as inputs to a classifier, such as an RBF neural network. Those attributes that increase the validation error are deemed irrelevant and are deleted. The complexity of the classifier can thus be reduced and its classification performance improved. Computer simulations show that our method for attribute importance ranking leads to smaller attribute subsets with higher accuracies compared with the existing SUD and Relief-F methods. We also propose a modified method for efficient construction of an RBF classifier. In this method we allow for large overlaps between clusters corresponding to the same class label. Our approach significantly reduces the structural complexity of the RBF network and improves the classification performance.

A noisy chaotic neural network for solving combinatorial optimization problems: stochastic chaotic simulated annealing

Recently Chen and Aihara have demonstrated both experimentally and mathematically that their chaotic simulated annealing (CSA) has better search ability for solving combinatorial optimization problems compared to both the Hopfield-Tank approach and stochastic simulated annealing (SSA). However, CSA may not find a globally optimal solution no matter how slowly annealing is carried out, because the chaotic dynamics are completely deterministic. In contrast, SSA tends to settle down to a global optimum if the temperature is reduced sufficiently slowly. Here we combine the best features of both SSA and CSA, thereby proposing a new approach for solving optimization problems, i.e., stochastic chaotic simulated annealing, by using a noisy chaotic neural network. We show the effectiveness of this new approach with two difficult combinatorial optimization problems, i.e., a traveling salesman problem and a channel assignment problem for cellular mobile communications.

Rule extraction by genetic algorithms based on a simplified RBF neural network

As an important task of data mining, extracting rules to represent the concept of numerical data is attracting much attention. We propose a novel algorithm to extract rules using genetic algorithms (GA) and the radial basis function (RBF) neural network classifier. The interval for each input in the condition part of each rule is adjusted using GA. The fitness of a chromosome is determined by the accuracy of extracted rules. The decision boundary of rules extracted is hyper-rectangular. During the training of the RBF neural network, large overlaps between clusters corresponding to the same class is allowed in order to decrease the number of hidden units while maintaining classification accuracy. The weights connecting the hidden units with the output units are then pruned. Our simulations demonstrate that our approach leads to more accurate and concise rules.

A GA-based RBF classifier with class-dependent features

High dimensionality of data sets is a curse to classifiers. We propose to construct a novel radial basis function (RBF) classifier using class-dependent features by genetic algorithms (GA). Since each feature may have different capabilities in discriminating different classes, features should be masked differently for different classes. In our novel RBF classifier, each Gaussian kernel function of the RBF neural network is active for only a subset of patterns which are approximately of the same class. A group of Gaussian kernel functions is generated for each class. In our method, different feature masks are used for different groups of Gaussian kernel functions corresponding to different classes. The feature masks are adjusted by GA. The classification accuracy of the RBF neural network is used as the fitness function. Thus, the dimensionality of a data set is reduced. Simulations show that, with irrelevant features removed for each class, our method can lead to significant improvements on classification accuracy.

Epidemic reemergence in adaptive complex networks

The dynamic nature of a system gives rise to dynamical features of epidemic spreading, such as oscillation and bistability. In this paper, by studying the epidemic spreading in growing networks, in which susceptible nodes may adaptively break the connections with infected ones yet avoid being isolated, we reveal a phenomenon, epidemic reemergence, where the number of infected nodes is incubated at a low level for a long time and then erupts for a short time. The process may repeat several times before the infection finally vanishes. Simulation results show that all three factors, namely the network growth, the connection breaking, and the isolation avoidance, are necessary for epidemic reemergence to happen. We present a simple theoretical analysis to explain the process of reemergence in detail. Our study may offer some useful insights, helping explain the phenomenon of repeated epidemic explosions.

Statistical modeling reveals the effect of absolute humidity on dengue in Singapore.

Weather factors are widely studied for their effects on indicating dengue incidence trends. However, these studies have been limited due to the complex epidemiology of dengue, which involves dynamic interplay of multiple factors such as herd immunity within a population, distinct serotypes of the virus, environmental factors and intervention programs. In this study, we investigate the impact of weather factors on dengue in Singapore, considering the disease epidemiology and profile of virus serotypes. A Poisson regression combined with Distributed Lag Non-linear Model (DLNM) was used to evaluate and compare the impact of weekly Absolute Humidity (AH) and other weather factors (mean temperature, minimum temperature, maximum temperature, rainfall, relative humidity and wind speed) on dengue incidence from 2001 to 2009. The same analysis was also performed on three sub-periods, defined by predominant circulating serotypes. The performance of DLNM regression models were then evaluated through the Akaikes Information Criterion. From the correlation and DLNM regression modeling analyses of the studied period, AH was found to be a better predictor for modeling dengue incidence than the other unique weather variables. Whilst mean temperature (MeanT) also showed significant correlation with dengue incidence, the relationship between AH or MeanT and dengue incidence, however, varied in the three sub-periods. Our results showed that AH had a more stable impact on dengue incidence than temperature when virological factors were taken into consideration. AH appeared to be the most consistent factor in modeling dengue incidence in Singapore. Considering the changes in dominant serotypes, the improvements in vector control programs and the inconsistent weather patterns observed in the sub-periods, the impact of weather on dengue is modulated by these other factors. Future studies on the impact of climate change on dengue need to take all the other contributing factors into consideration in order to make meaningful public policy recommendations.

Data dimensionality reduction with application to improving classification performance and explaining concepts of data sets

Data dimensionality reduction is usually carried out before patterns are input to classifiers. In order to obtain good results in data mining, selecting relevant data is desirable. In many cases, irrelevant or redundant attributes are included in data sets, which interfere with knowledge discovery from data sets. In this paper, we propose a rule-extraction method based on a novel separability-correlation measure (SCM) ranking the importance of attributes. According to the attribute ranking results, the attribute subsets that lead to the best classification results are selected and used as inputs to a classifier, such as an RBF neural network in our paper. The complexity of the classifier can thus be reduced and its classification performance improved. Our method uses the classification results with reduced attribute sets to extract rules. Computer simulations show that our method leads to smaller rule sets with higher accuracies compared with other methods.

Training RBF neural networks on unbalanced data

This paper presents a new algorithm for the construction and training of an RBF neural network with unbalanced data. In applications, minority classes with much fewer samples are often present in data sets. The learning process of a neural network usually is biased towards classes with majority populations. Our study focused on improving the classification accuracy of minority classes while maintaining the overall classification performance.

Evaluating temporal factors in combined interventions of workforce shift and school closure for mitigating the spread of influenza.

Background It is believed that combined interventions may be more effective than individual interventions in mitigating epidemic. However there is a lack of quantitative studies on performance of the combination of individual interventions under different temporal settings. Methodology/Principal Findings To better understand the problem, we develop an individual-based simulation model running on top of contact networks based on real-life contact data in Singapore. We model and evaluate the spread of influenza epidemic with intervention strategies of workforce shift and its combination with school closure, and examine the impacts of temporal factors, namely the trigger threshold and the duration of an intervention. By comparing simulation results for intervention scenarios with different temporal factors, we find that combined interventions do not always outperform individual interventions and are more effective only when the duration is longer than 6 weeks or school closure is triggered at the 5% threshold; combined interventions may be more effective if school closure starts first when the duration is less than 4 weeks or workforce shift starts first when the duration is longer than 4 weeks. Conclusions/Significance We therefore conclude that identifying the appropriate timing configuration is crucial for achieving optimal or near optimal performance in mitigating the spread of influenza epidemic. The results of this study are useful to policy makers in deliberating and planning individual and combined interventions.

Robustness of scale-free networks under rewiring operations

Scale-free networks have strong tolerance against random failures yet are fragile under intentional attacks. Existing results show that the network robustness can also be affected by its correlation profile. Specifically, scale-free networks with larger assortativity coefficients generally tend to be more robust against intentional attack. In this letter, we reveal some interesting different observations. By proposing a simple rewiring method which does not change any nodal degree, we show that network robustness can be steadily enhanced at a slightly decreased assortativity coefficient. The tolerance against random failures meanwhile remains largely unaffected. Such observations demonstrate the more complicated relationship between network robustness and its assortativity level, as well as some new possibilities of network enhancement and protection.