The need for low bias algorithms in classification learning from large data sets

This paper reviews the appropriateness for application to large data sets of standard machine learning algorithms, which were mainly developed in the context of small data sets. Sampling and parallelisation have proved useful means for reducing computation time when learning from large data sets. However, such methods assume that algorithms that were designed for use with what are now considered small data sets are also fundamentally suitable for large data sets. It is plausible that optimal learning from large data sets requires a different type of algorithm to optimal learning from small data sets. This paper investigates one respect in which data set size may affect the requirements of a learning algorithm — the bias plus variance decomposition of classification error. Experiments show that learning from large data sets may be more effective when using an algorithm that places greater emphasis on bias management, rather than variance management.

Missing value estimation for mixed-attribute data sets

Missing data imputation is a key issue in learning from incomplete data. Various techniques have been developed with great successes on dealing with missing values in data sets with homogeneous attributes (their independent attributes are all either continuous or discrete). This paper studies a new setting of missing data imputation, i.e., imputing missing data in data sets with heterogeneous attributes (their independent attributes are of different types), referred to as imputing mixed-attribute data sets. Although many real applications are in this setting, there is no estimator designed for imputing mixed-attribute data sets. This paper first proposes two consistent estimators for discrete and continuous missing target values, respectively. And then, a mixture-kernel-based iterative estimator is advocated to impute mixed-attribute data sets. The proposed method is evaluated with extensive experiments compared with some typical algorithms, and the result demonstrates that the proposed approach is better than these existing imputation methods in terms of classification accuracy and root mean square error (RMSE) at different missing ratios.

A study and a directory of energy consumption data sets of buildings

Energy consumption data are required to perform analysis, modelling, evaluation, and optimisation of energy usage in buildings. While a variety of energy consumption data sets have been examined and reported in the literature, there is a lack of a comprehensive categorisation and analysis of the available data sets. In this study, an overview of energy consumption data of buildings is provided. Three common strategies for generating energy consumption data, i.e., measurement, survey, and simulation, are described. A number of important characteristics pertaining to each strategy and the resulting data sets are discussed. In addition, a directory of energy consumption data sets of buildings is developed. The data sets are collected from either published papers or energy related organisations. The main contributions of this study include establishing a resource pertaining to energy consumption data sets and providing information related to the characteristics and availability of the respective data sets; therefore facilitating and promoting research activities in energy consumption data analysis.

A highly practical approach toward achieving minimum data sets storage cost in the cloud

Massive computation power and storage capacity of cloud computing systems allow scientists to deploy computation and data intensive applications without infrastructure investment, where large application data sets can be stored in the cloud. Based on the pay-as-you-go model, storage strategies and benchmarking approaches have been developed for cost-effectively storing large volume of generated application data sets in the cloud. However, they are either insufficiently cost-effective for the storage or impractical to be used at runtime. In this paper, toward achieving the minimum cost benchmark, we propose a novel highly cost-effective and practical storage strategy that can automatically decide whether a generated data set should be stored or not at runtime in the cloud. The main focus of this strategy is the local-optimization for the tradeoff between computation and storage, while secondarily also taking users' (optional) preferences on storage into consideration. Both theoretical analysis and simulations conducted on general (random) data sets as well as specific real world applications with Amazon's cost model show that the cost-effectiveness of our strategy is close to or even the same as the minimum cost benchmark, and the efficiency is very high for practical runtime utilization in the cloud.

SCLOPE: an algorithm for clustering data streams of categorical attributes

Clustering is a difficult problem especially when we consider the task in the context of a data stream of categorical attributes. In this paper, we propose SCLOPE, a novel algorithm based on CLOPErsquos intuitive observation about cluster histograms. Unlike CLOPE however, our algo- rithm is very fast and operates within the constraints of a data stream environment. In particular, we designed SCLOPE according to the recent CluStream framework. Our evaluation of SCLOPE shows very promising results. It consistently outperforms CLOPE in speed and scalability tests on our data sets while maintaining high cluster purity; it also supports cluster analysis that other algorithms in its class do not.

Calculating minimum sample sizes for taxonomic measurements : examples using Gäumann’s Peronospora Spore Data

Methods are presented for calculating minimum sample sizes necessary to obtain precise estimates of fungal spore dimensions. Using previously published spore-length data sets for Peronospora species, we demonstrate that 41—71 spores need to be measured to estimate the mean length with a reasonable level of statistical precision and resolution. This is further progressed with examples for calculating the minimum number of spore lengths to measure when matching an undetermined specimen to a known species. Although applied only to spore-length data, all described methods can be applied to any morphometric data that satisfy certain statistical assumptions.

Discovering linear causal model from incomplete data

One common drawback in algorithms for learning Linear Causal Models is that they can not deal with incomplete data set. This is unfortunate since many real problems involve missing data or even hidden variable. In this paper, based on multiple imputation, we propose a three-step process to learn linear causal models from incomplete data set. Experimental results indicate that this algorithm is better than the single imputation method (EM algorithm) and the simple list deletion method, and for lower missing rate, this algorithm can even find models better than the results from the greedy learning algorithm MLGS working in a complete data set. In addition, the method is amenable to parallel or distributed processing, which is an important characteristic for data mining in large data sets.

CIDB: Chlamydia Interactive Database for cross-querying genomics, transcriptomics and proteomics data

Chlamydiae are important pathogens of humans, birds and a wide range of animals. They are a unique group of bacteria, characterized by their developmental cycle. Chlamydia has been difficult to study because of their obligate intracellular growth habit and lack of a genetic transformation system. However, the past 5 years has seen the full genome sequencing of seven strains of Chlamydia and a rapid expansion of genomic, transcriptomic (RT-PCR, microarray) and proteomic analysis of these pathogens. The Chlamydia Interactive Database (CIDB) described here is the first database of its type that holds genomic, RT-PCR, microarray and proteomics data sets that can be cross-queried by researchers for patterns in the data. Combining the data of many research groups into a single database and cross-querying from different perspectives should enhance our understanding of the complex cell biology of these pathogens. The database is available at: http://www3.it.deakin.edu.au:8080/CIDB/.

Estimating mean exposures from censored data: exposure to benzene in the Australian petroleum industry

A retrospective assessment of exposure to benzene was carried out for a nested case control study of lympho-haematopoietic cancers, including leukaemia, in the Australian petroleum industry. Each job or task in the industry was assigned a Base Estimate (BE) of exposure derived from task-based personal exposure assessments carried out by the company occupational hygienists. The BEs corresponded to the estimated arithmetic mean exposure to benzene for each job or task and were used in a deterministic algorithm to estimate the exposure of subjects in the study. Nearly all of the data sets underlying the BEs were found to contain some values below the limit of detection (LOD) of the sampling and analytical methods and some were very heavily censored; up to 95% of the data were below the LOD in some data sets. It was necessary, therefore, to use a method of calculating the arithmetic mean exposures that took into account the censored data. Three different methods were employed in an attempt to select the most appropriate method for the particular data in the study. A common method is to replace the missing (censored) values with half the detection limit. This method has been recommended for data sets where much of the data are below the limit of detection or where the data are highly skewed; with a geometric standard deviation of 3 or more. Another method, involving replacing the censored data with the limit of detection divided by the square root of 2, has been recommended when relatively few data are below the detection limit or where data are not highly skewed. A third method that was examined is Cohen's method. This involves mathematical extrapolation of the left-hand tail of the distribution, based on the distribution of the uncensored data, and calculation of the maximum likelihood estimate of the arithmetic mean. When these three methods were applied to the data in this study it was found that the first two simple methods give similar results in most cases. Cohen's method on the other hand, gave results that were generally, but not always, higher than simpler methods and in some cases gave extremely high and even implausible estimates of the mean. It appears that if the data deviate substantially from a simple log-normal distribution, particularly if high outliers are present, then Cohen's method produces erratic and unreliable estimates. After examining these results, and both the distributions and proportions of censored data, it was decided that the half limit of detection method was most suitable in this particular study.

Managing data using neighbor replication on triangular-grid structure

Data is one of the domains in grid research that deals with the storage, replication, and management of large data sets in a distributed environment. The all-data-to-all sites replication scheme such as read-one write-all and tree grid structure (TGS) are the popular techniques being used for replication and management of data in this domain. However, these techniques have its weaknesses in terms of data storage capacity and also data access times due to some number of sites must ‘agree’ in common to execute certain transactions. In this paper, we propose the all-data-to-some-sites scheme called the neighbor replication on triangular grid (NRTG) technique by considering only neighbors have the replicated data, and thus, minimizes the storage capacity as well as high update availability. Also, the technique tolerates failures such as server failures, site failure or even network partitioning using remote procedure call (RPC).

Bayesian classification of catchments using spatial data: a first step to improved modelling of catchment effects on stream ecological condition

A major challenge facing freshwater ecologists and managers is the development of models that link stream ecological condition to catchment scale effects, such as land use. Previous attempts to make such models have followed two general approaches. The bottom-up approach employs mechanistic models, which can quickly become too complex to be useful. The top-down approach employs empirical models derived from large data sets, and has often suffered from large amounts of unexplained variation in stream condition.

We believe that the lack of success of both modelling approaches may be at least partly explained by scientists considering too wide a breadth of catchment type. Thus, we believe that by stratifying large sets of catchments into groups of similar types prior to modelling, both types of models may be improved. This paper describes preliminary work using a Bayesian classification software package, ‘Autoclass’ (Cheeseman and Stutz 1996) to create classes of catchments within the Murray Darling Basin based on physiographic data.

Autoclass uses a model-based classification method that employs finite mixture modelling and trades off model fit versus complexity, leading to a parsimonious solution. The software provides information on the posterior probability that the classification is ‘correct’ and also probabilities for alternative classifications. The importance of each attribute in defining the individual classes is calculated and presented, assisting description of the classes. Each case is ‘assigned’ to a class based on membership probability, but the probability of membership of other classes is also provided. This feature deals very well with cases that do not fit neatly into a larger class. Lastly, Autoclass requires the user to specify the measurement error of continuous variables.

Catchments were derived from the Australian digital elevation model. Physiographic data werederived from national spatial data sets. There was very little information on measurement errors for the spatial data, and so a conservative error of 5% of data range was adopted for all continuous attributes. The incorporation of uncertainty into spatial data sets remains a research challenge.

The results of the classification were very encouraging. The software found nine classes of catchments in the Murray Darling Basin. The classes grouped together geographically, and followed altitude and latitude gradients, despite the fact that these variables were not included in the classification. Descriptions of the classes reveal very different physiographic environments, ranging from dry and flat catchments (i.e. lowlands), through to wet and hilly catchments (i.e. mountainous areas). Rainfall and slope were two important discriminators between classes. These two attributes, in particular, will affect the ways in which the stream interacts with the catchment, and can thus be expected to modify the effects of land use change on ecological condition. Thus, realistic models of the effects of land use change on streams would differ between the different types of catchments, and sound management practices will differ.

A small number of catchments were assigned to their primary class with relatively low probability. These catchments lie on the boundaries of groups of catchments, with the second most likely class being an adjacent group. The locations of these ‘uncertain’ catchments show that the Bayesian classification dealt well with cases that do not fit neatly into larger classes.

Although the results are intuitive, we cannot yet assess whether the classifications described in this paper would assist the modelling of catchment scale effects on stream ecological condition. It is most likely that catchment classification and modelling will be an iterative process, where the needs of the model are used to guide classification, and the results of classifications used to suggest further refinements to models.

Data pre-processing for more effective gene clustering

The high-throughput experimental data from the new gene microarray technology has spurred numerous efforts to find effective ways of processing microarray data for revealing real biological relationships among genes. This work proposes an innovative data pre-processing approach to identify noise data in the data sets and eliminate or reduce the impact of the noise data on gene clustering, With the proposed algorithm, the pre-processed data sets make the clustering results stable across clustering algorithms with different similarity metrics, the important information of genes and features is kept, and the clustering quality is improved. The primary evaluation on real microarray data sets has shown the effectiveness of the proposed algorithm.