Entity

Time filter

Source Type

Johannesburg, South Africa

Duma M.,University of Johannesburg | Marwala T.,University of Johannesburg | Twala B.,University of Johannesburg | Nelwamondo F.,Modelling and Digital Science Unit
Applied Soft Computing Journal | Year: 2013

Missing data in large insurance datasets affects the learning and classification accuracies in predictive modelling. Insurance datasets will continue to increase in size as more variables are added to aid in managing client risk and will therefore be even more vulnerable to missing data. This paper proposes a hybrid multi-layered artificial immune system and genetic algorithm for partial imputation of missing data in datasets with numerous variables. The multi-layered artificial immune system creates and stores antibodies that bind to and annihilate an antigen. The genetic algorithm optimises the learning process of a stimulated antibody. The evaluation of the imputation is performed using the RIPPER, k-nearest neighbour, naïve Bayes and logistic discriminant classifiers. The effect of the imputation on the classifiers is compared with that of the mean/mode and hot deck imputation methods. The results demonstrate that when missing data imputation is performed using the proposed hybrid method, the classification improves and the robustness to the amount of missing data is increased relative to the mean/mode method for data missing completely at random (MCAR) missing at random (MAR), and not missing at random (NMAR).The imputation performance is similar to or marginally better than that of the hot deck imputation. © 2013 Elsevier B.V. All rights reserved. Source


Nelwamondo F.V.,Modelling and Digital Science Unit | Nelwamondo F.V.,University of Johannesburg | Golding D.,Modelling and Digital Science Unit | Marwala T.,University of Johannesburg
Information Sciences | Year: 2013

This paper develops and presents a novel technique for missing data estimation using a combination of dynamic programming, neural networks and genetic algorithms (GA) on suitable subsets of the input data. The method proposed here is well suited for decision making processes and uses the concept of optimality and the Bellman's equation to estimate the missing data. The proposed approach is applied to an HIV/AIDS database and the results shows that the proposed method significantly outperforms a similar method where dynamic programming is not used. This paper also suggests a different way of formulating a missing data problem such that the dynamic programming is applicable to estimate the missing data. © 2013 Elsevier Inc. All rights reserved. Source


Moolla Y.,University of KwaZulu - Natal | Moolla Y.,Modelling and Digital Science Unit | Viriri S.,University of KwaZulu - Natal | Nelwamondo F.V.,Modelling and Digital Science Unit | Tapamo J.R.,University of KwaZulu - Natal
2012 IEEE International Conference on Signal Processing, Communications and Computing, ICSPCC 2012 | Year: 2012

Signatures are one of the behavioural biometric traits, which are widely used as a means of personal verification. Therefore, they require efficient and accurate methods of authenticating users. The use of a single distance-based classification technique normally results in a lower accuracy compared to supervised learning techniques. This paper investigates the use of a combination of multiple distance-based classification techniques, namely individually optimized re-sampling, weighted Euclidean distance, fractional distance and weighted fractional distance. Results are compared to a similar system that uses support vector machines. It is shown that competitive levels of accuracy can be obtained using distance-based classification. The best accuracy obtained is 89.2%. © 2012 IEEE. Source


Nelufule N.,Modelling and Digital Science Unit | Nelwamondo F.,Modelling and Digital Science Unit | Nelwamondo F.,University of Johannesburg | Malumedzha T.,Modelling and Digital Science Unit | Marwala T.,University of Johannesburg
International Journal of Innovative Computing, Information and Control | Year: 2015

In this paper we propose an adaptive fusion approach for iris biometric. The proposed fusion method incorporates four matching algorithms using feature quality and relative entropy to enhance iris fusion performance. This method introduces relative entropy measure to the fusion process to assign low weighting coefficients to features with less information and higher weights to features with more information. We investigated the parameters which influence the rejection rates and acceptance rates to determine the optimal equal error rate. The best equal error rates were aimed at high recognition accuracy. The proposed method was tested on two public iris databases. CASIA left eye images produced 99.36% recognition accuracy and 0.041% equal error rate as compared to 98.93% recognition accuracy and 0.066% error rate produced by the weighted sum fusion. For the CASIA right eye images, the proposed method produced 99.18% recognition accuracy and 0.087% equal error rate as compared to weighted sum fusion with 98.81% recognition accuracy and 0.096% equal error rate. From the UBIRIS database, the proposed method produced 99.59% recognition accuracy and 0.038% equal error rate as compared to 98.53% recognition accuracy and 0.074% equal error rate produced by weighted sum fusion. The proposed method shows improved recognition performance in terms of AUC and the EER. © 2015 ICIC International. Source

Discover hidden collaborations