2022-03

2022 International Conference on Computer Science and Software Engineering (CSASE)

The redundant or irrelevant features in microarray datasets cause difficulty in apprehending the prospect patterns directly and accurately. One of the necessary strategies for distinguishing and screening out the most relevant features is Feature Selection (FS). However, the increasing feature dimensions and small sample size in microarray datasets pose a significant challenge to most existing algorithms. To overcome this issue, we propose a novel method based on Principle Component Analysis (PCA) and Cuttlefish Algorithm (CFA), which is a recent bio-inspired feature selection algorithm. The critical characteristic of the PCA algorithm is that it is less sensitive to noise and requires less memory and capacity. Furthermore, adopting the PCA approach before using CFA minimises the search space within CFA, which speeds up determining the best subset of features while reducing the computational cost. To assess the performance of the proposed method, three publicly available microarray datasets are utilized in the experimental studies using a Linear Discriminant Analysis classifier. Experimental results showed that PCA with CFA significantly outperforms the state-of-art feature selection methods.

2022-03

2022 International Conference on Computer Science and Software Engineering (CSASE)

The Internet of Things (IoT) has attracted significant attention from researchers and companies as it covers a wide range of applications, including industrial control, healthcare, transportation, smart cities and homes, and agriculture. As the pioneering IoT application, smart homes aim to increase the quality of residents' lives by making home appliances automated by remotely controlling or automatically operating them far from home. However, not only remotely controlling should take into consideration when building a smart home system, but also making the home environment as smart as possible by incorporating the value and power of Machine Learning (ML) techniques to it. Adding such value is crucial to make the system adaptive to users' activities toward predicting user behavior to reduce power consumption further, enhance security levels, and improve usability experiences. This study reviews popular machine learning techniques in smart home applications with their benefits and limitations. The paper also provides a comprehensive comparison among the available systems in the literature which integrate smart homes with the intelligence of the ML algorithms. Finally, after discussing the results, the open problems and suggested directions are presented, and the challenges and perspectives of future development of smart homes systems are presented and discussed.

2019-10

4. International Conference on Theoretical and Applied Computer Science and Engineering 2019

Diabetes is a widely distributed disease in the world. It is a chronic disease characterized by hyperglycemia. This disease causes various complications and high mortality and morbidity rate. Early diagnose of diabetes is very crucial for timely treatment. There are different ways to detect the diabetes; one of which is the use of machine learning algorithms. Machine learning techniques have been widely used in medical health field to diagnose and predict the occurrences of diseases. In this study, six algorithms, such as Logistic Regression (LR), Decision Tree (DT), K-Nearest Neighbors (K-NN), Naïve Bayes (NB), Support Vector Machine (SVM) and Artificial Neural Networks (ANN) were used to diagnose and classify type II diabetes of Pima dataset. However, classes are not always balanced, and imbalanced data occurs when one of the classes is minority and the other one is majority. The main issue of imbalanced data is usually resulted in misclassification where the minority class tends to be misclassified. To overcome this problem, different methods can be used. In this paper, we investigated the use of SMOTE resampling method to balance the data, and the results of the six algorithms were compared for balanced and imbalanced data. Experimental results showed that classification with resampling/ balancing has significant improvement up to 20% of accuracy for some models.

2018-10

11. International Statistics Days Conference, 3 - 7 October 2018, TURKEY

A hierarchical Bayesian approach for dating tree-ring sequences is introduced and demonstrated in a relatively small but computationally challenging exercise. The Bayesian inference via two-stage Markov chain Monte Carlo includes evaluating the likelihood of model parameters at every possible offset, which allows the posterior distribution of the unknown date to be estimated. Implementation involves generating pseudo-chronologies from a mechanistic model, called VSLite, and matching undated sequences to generated pseudo-chronologies thus providing posterior estimates of the match. Results show that the approach has successfully matched undated sequences to the pseudo-master chronology and hence provided useful information about the most likely offset.

2018-10

International Conference on Advanced Science and Engineering 2018 (ICOASE2018)

A nonlinear forward model named VSLite is used to simulate tree ring-width growth from climate data. There is always uncertainty in such data inputs, which might influence the uncertainty of the model outputs. The present work performs a Bayesian sensitivity analysis (BSA) to the VSLite model using a Gaussian process emulator. BSA aims to understand and quantify the uncertainty of the model’s outputs due to a change in its inputs. The model was successfully implemented at different geographical locations around the world. To examine the accuracy of the model, we first compared real tree-ring data at different locations with those simulated from VSLite. The variability in the model output was then explored and quantified via BSA. Results show that BSA has successfully classified model parameters in terms of their influences on the model output variation.

2014-09

Second Bayesian Young Statistician Meeting, Vienna (September 2014).

A nonlinear forward model, VSLite, simulates tree ring widths using climate data as inputs. There is uncertainty in the inputs which in uence the model outputs. Initial comparison of the distributions of inter-annual ring widths from real trees with those simulated by VSLite, suggests that the latter exhibits considerably more variability than the former. In order to explore this, we conducted a Bayesian sensitivity analysis (BSA) of VSLite using a Gaussian process emulator. Such an analysis allows us to understand and quantify the uncertainty of the model's outputs due to changes in its inputs. In this talk we will report on our experiments and explore why and how the observed differences occur.