Nor Idayu Mahat

Variable selection in discriminant analysis based on the location model for mixed variables

Non-parametric smoothing of the location model is a potential basis for discriminating between groups of objects using mixtures of continuous and categorical variables simultaneously. However, it may lead to unreliable estimates of parameters when too many variables are involved. This paper proposes a method for performing variable selection on the basis of distance between groups as measured by smoothed Kullback–Leibler divergence. Searching strategies using forward, backward and stepwise selections are outlined, and corresponding stopping rules derived from asymptotic distributional results are proposed. Results from a Monte Carlo study demonstrate the feasibility of the method. Examples on real data show that the method is generally competitive with, and sometimes is better than, other existing classification methods.

Principal Component Analysis – A Realization of Classification Success in Multi Sensor Data Fusion

The field of measurement technology in the sensors domain is rapidly changing due to the availability of statistical tools to handle many variables simultaneously. The phenomenon has led to a change in the approach of generating dataset from sensors. Nowadays, multiple sensors, or more specifically multi sensor data fusion (MSDF) are more favourable than a single sensor due to significant advantages over single source data and has better presentation of real cases. MSDF is an evolving technique related to the problem for combining data systematically from one or multiple (and possibly diverse) sensors in order to make inferences about a physical event, activity or situation. Mitchell (2007) defined MSDF as the theory, techniques, and tools which are used for combining sensor data, or data derived from sensory data into a common representational format. The definition also includes multiple measurements produced at different time instants by a single sensor as described by (Smith & Erickson, 1991).

Understanding Mahalanobis distance criterion for feature selection

Distance criteria are widely applied in cluster analysis and classification techniques. One of the well known and most commonly used distance criteria is the Mahalanobis distance, introduced by P. C. Mahalanobis in 1936. The functions of this distance have been extended to different problems such as detection of multivariate outliers, multivariate statistical testing, and class prediction problems. In the class prediction problems, researcher is usually burdened with problems of excessive features where useful and useless features are all drawn for classification task. Therefore, this paper tries to highlight the procedure of exploiting this criterion in selecting the best features for further classification process. Classification performance for the feature subsets of the ordered features based on the Mahalanobis distance criterion is included.

Investigation on zakat as an indicator for Moslem countries' economic growth

This study examines the possibility of zakat as an alternative source of economic growth, which in most cases and prior studies Foreign Direct Investment (FDI) has been more preferable and become dominant development engine for emerging economies. We proposed the strategy to estimate the zakat and employed a multivariate regression to analyse a panel macroeconomic data of 19 Moslem countries for the period 2004–2010. The empirical results give some evidence that zakat is able to become a powerful resource of economic growth, or in other words, it could be domestic direct investment that provides source of fund to the country development. Although the proposed model is limited due to the availability of complete panel macroeconomic data, we conclude that zakat will become a serious implementable economic growth policy in Moslem countries reflecting a self-sufficient spirit, equitable distribution of wealth paradigm, and much more human-touch in managing national development.

Modeling of spectrum demands through hybrids of analytic hierarchy process and integer programming

As wireless technology advances, demand for spectrum bandwidth increases and thus, spectrum has become scarce. As a scarce resource, spectrum bandwidths need to be efficiently allocated to potential service providers. Hence, this paper first reviews current approaches on how spectrum bandwidths are being allocated to the service providers. Then we present an efficient integrated approach in allocating spectrum volumes, whereby the approaches of Analytic Hierarchy Process and Integer Programming are integrated and applied to produce systematic and consistent allocation results. The integrated approach is able to cater multi-criteria problems through determination of suitable weights and computations, which exhibits a more efficient alternative as compared to the existing approaches.The illustrations revealed that the integrated approach indeed has the potential to be implemented and gives an alternative to the current existing approaches.

The stopping rules for winsorized tree

Winsorized tree is a modified tree-based classifier that is able to investigate and to handle all outliers in all nodes along the process of constructing the tree. It overcomes the tedious process of constructing a classical tree where the splitting of branches and pruning go concurrently so that the constructed tree would not grow bushy. This mechanism is controlled by the proposed algorithm. In winsorized tree, data are screened for identifying outlier. If outlier is detected, the value is neutralized using winsorize approach. Both outlier identification and value neutralization are executed recursively in every node until predetermined stopping criterion is met. The aim of this paper is to search for significant stopping criterion to stop the tree from further splitting before overfitting. The result obtained from the conducted experiment on pima indian dataset proved that the node could produce the final successor nodes (leaves) when it has achieved the range of 70% in information gain.

Comparisons of parametric and non-parametric classification rules for e-nose and e-tongue

This paper evaluates the performance of parametric and non-parametric classification rules in sensor technology. The growing of sensor technologies, e-nose and e-tongue, has urged engineers to equip themselves with the utmost recent and advanced statistical approaches. As data collected from e-nose and e-tongue face some complexities, often data pre-processing and transformation are performed prior to the classification. This paper discusses the comparisons made on some known parametric and non-parametric classification rules in the application for classifying data of e-nose and e-tongue. The comparisons which based on leave-one-out accuracy, sensitivity and specificity shows that non-parametric approaches especially k-nearest neighbour does not much distorted with changes of distribution, but Naive Bayes is greatly influenced by the structure of the data.

Sensors closeness test based on an improved [0, 1] bounded Mahalanobis distance Δ2

Mahalanobis distance Δ2 values are commonly in the range of 0 to +∞ where higher values represent greater distance between class means or points. The increase in Mahalanobis distance is unbounded as the distance multiply. To certain extend, the unbounded distance values pose difficulties in the evaluation and decision for instance in the sensors closeness test. This paper proposes an approach to [0, 1] bounded Mahalanobis distance Δ2 that enable researcher to easily perform sensors closeness test. The experimental data of four different types of rice based on three different electronic nose sensors namely InSniff, PEN3, and Cyranose320 were analyzed and sensor closeness test seems successfully performed within the [0, 1] bound.

Integrating analytic hierarchy process with integer programming to efficiently model spectrum demands

As wireless technology advances, demand for spectrum bandwidth increases and thus, spectrum has become scarce. As a scarce resource, spectrum bandwidths need to be efficiently allocated to potential service providers. Hence, this paper presents an efficient integrated approach in allocating spectrum volumes, whereby the approaches of Analytic Hierarchy Process and Integer Programming are combined and applied to produce systematic and consistent results. The integrated approach is able to cater multi-criteria problems through determination of suitable weights and computations, which exhibits a more efficient alternative as compared to the existing approaches.