University of Technology Malaysia
Universiti Teknologi Malaysia Institutional RepositoryNot a member yet
70456 research outputs found
Sort by
Enhancing extreme learning machines using cross-entropy moth-flame optimization algorithm
No full textExtreme Learning Machines (ELM) learn fast and eliminate the tuning of input weights and biases. However, ELM does not guarantee the optimal setting of the weights and biases due to random input parameters initialization. Therefore, ELM suffers from instability of output, large network size, and degrade generalization performance. To overcome these problems, an efficient co-evolutionary hybrid model namely as Cross-Entropy Moth-Flame Optimization (CEMFO-ELM) model is proposed to train a neural network for the selection of optimal input weights and biases. The hybrid model balanced the exploration and exploitation of the search space, and then selected optimal input weights and biases for ELM. The co-evolutionary algorithm reduced the chances of been trapped into the local extremum in the search space. Accuracy, stability, and percentage improvement ratio (PIR%) were the metrics used to evaluate the performance of the proposed model when simulated on some classification datasets for machine learning from the University of California, Irvine repository. The co-evolutionary scheme was compared with its constituent individual ELM-based enhanced meta-heuristic schemes (CE-ELM and MFO-ELM). The co-evolutionary meta-heuristic algorithm enhances the selection of optimal parameters for ELM. It improves the accuracy of ELM in all the simulations, and the stability of ELM was improved in all, up to 53% in Breast cancer simulation. Also, it has better convergences than the comparative ELM hybrid model in all the simulations
Perceptions of returnees concerning their rehabilitation and reinstatement in Swat District, Pakistan: an evaluative study
Get PDFThis article examines the perception of returnees about their reinstatement and rehabilitation in the Swat District of Pakistan. A satisfaction tool, consisting of various domains and indicators, was used for measuring the returnee’s perceptions at two periods, i.e., before rehabilitation (BR) and after rehabilitation (AR). Data were elicited through a self-administered structured questionnaire from 382 samples drawn from the 47,943 Kabal Tehsil, Swat population. Data were analyzed through descriptive and inferential statistics. The findings depict that the value of all the domains increased by 25.7%. The paired sample t-test results show a rejection of all the null hypotheses, indicating a significant increase in the overall satisfaction of returnees in the AR period. The findings indicate that in the AR period, the highest increase occurred in SWL (Satisfaction with Life) domain and the lowest in GOV (Government). This study concludes that the satisfaction of returnees can be further improved by focusing on the domains with a lower level of satisfaction, such as the Government and Social Support domains. Additionally, the tool adopted in this study is significant for measuring the satisfaction level of the distressed population in Pakistan and beyond
Die-level defects classification using region-based convolutional neural network
No full textVisual inspection process on semiconductors is usually performed by human experts. These inspection tasks require extreme concentration, and the time an inspector could continue the inspection tasks is limited. An automated die-level defects classification system is presented in this paper to replace human experts in inspection tasks. The proposed system utilizes a Region-based Convolutional Neural Network on die-level images for defect detection and classification. Four defect classes are considered, blob, die crack, pinhole, and underfill. The proposed method achieved 88.5% and 71.4% accuracy in detection and defect classification, which is equivalent to that performed by human inspectors of between 60 - 80%
University examination timetabling using a hybrid black hole algorithm
No full textUniversity timetabling construction is a complicated task that universities worldwide encounter. This study developed a hybrid approach to produce a timetable solution for the university examination timetabling problem. Black Hole Algorithm (BHA), a population-based approach that mimics the black hole phenomenon, has recently been introduced in the literature and successfully applied in addressing various optimization problems. Although its effectiveness has been proven, there still exists inefficiency regarding the exploitation ability where BHA is poor in fine-tuning search region in reaching for good quality of the solution. Hence, a hybrid framework for university examination timetabling problem that is based on BHA and Hill Climbing local search is proposed (hybrid BHA). This hybridization aims to improve the exploitation ability of BHA in fine-tuning the promising search regions and convergence speed of the search process. A real-world university examination benchmark dataset has been used to evaluate the performance of hybrid BHA. The computational results demonstrate that hybrid BHA can generate competitive results and record the best results for three instances compared to the reference approaches and current best-known recorded in the literature. Besides, findings from the Friedman tests show that the hybrid BHA ranked second and third in comparison with hybrid and meta-heuristic approaches (a total of 27 approaches) reported in the literature, respectively
Development of Machine Vision System for Riverine Debris Counting
No full textIn Malaysia, about 80% of freshwater sources come from rivers, but 44% of rivers are polluted. One of the river cleaning efforts is via Ocean Cleanup's Interceptor river cleaning machine. The efficiency depends on its location at the river, which is highly dependent on debris count along the river currently counted by human manual operators. Unfortunately, the process is not continuous and can only be done few hours in daylight. This project proposed to replace manual counting with a continuous automated debris counting system using computer vision. The system consists of a camera connected to a computer with algorithms that process the river live video feed and automatically detect and count riverine debris. The system was trained using three datasets over two You Only Look Once (YOLOv4) configurations producing six YOLOv4 models. The system was tested on a 5-minutes video of a flowing water source with floating debris, and the system's best performance, to match human counting, was by 110% or 10% better than human counting. This count may assist decision-making in locating the river cleaning interceptor and increase the efficiency of river cleaning activities
Prediction of student’s academic performance during online learning based on regression in support vector machine
Get PDFSince the Movement Control Order (MCO) was adopted, all the universities have implemented and modified the principle of online learning and teaching in consequence of Covid-19. This situation has relatively affected the students’ academic performance. Therefore, this paper employs the regression method in Support Vector Machine (SVM) to investigate the prediction of students’ academic performance in online learning during the Covid-19 pandemic. The data was collected from undergraduate students of the Department of Mathematics, Faculty of Science and Mathematics, Sultan Idris Education University (UPSI). Students’ Cumulative Grade Point Average (CGPA) during online learning indicates their academic performance. The algorithm of Support Vector Machine (SVM) as a machine learning was employed to construct a prediction model of students’ academic performance., Two parameters, namely C (cost) and epsilon of the Support Vector Machine (SVM) algorithm should be identified first prior to further analysis. The best parameter C (cost) and epsilon in SVM regression are 4 and 0.8. The parameters then were used for four kernels, i.e., radial basis function kernel, linear kernel, polynomial kernel, and sigmoid kernel. from the findings, the finest type of kernel is the radial basis function kernel, with the lowest support vector value and the lowest Root Mean Square Error (RMSE) which are 27 and 0.2557. Based on the research, the results show that the pattern of prediction of students’ academic performance is similar to the current CGPA. Therefore, Support Vector Machine regression can predict students’ academic performance
Characterisation of a bacterium from Tebrau Strait and screening of microbial genomes for dehalogenases
No full textCurrent study was to investigate the presence of dehalogenase in the isolated bacterium from Selat Tebrau that can grow on 2,2-dichloropropionic acid (2,2-DCP is an active compound in herbicide Dalapon®). Strain RN1, a Gram-negative and rod in shape was tentatively identified as Enterobacter cancerogenus based on basic biochemical and the 16S rRNA gene analyses. The calculated cells doubling time were 5.29 hours based on growth of the bacterium in liquid minimal media with 10, 20 and 30 mM of 2,2-DCP, respectively. However, no growth was observed at 40 mM 2,2-DCP liquid minimal medium due to increase in 2,2-DCP toxicity. It was hypoth-esized that, strain RN1 produced dehalogenase(s) which merits a further study of the genomic data of the same genus and species available in the database. A putative dehalogenase, designated as DehRN was located in the published data of Enterobac-ter cancerogenus. Pairwise of DehRN amino acids with known dehalogenase re-sulted in sequence identity of <20%, suggesting a new class of dehalogenase enzyme in the Enterobacter cancerogenus
Polymer-free transfer of graphene-based material derived from cooking palm oil by chemical vapour deposition technique
No full textChemical vapour deposition (CVD) of cooking palm oil precursors with a nickel (Ni) catalyst is an established method to produce graphene-based materials. Nonetheless, transferring the graphene sheets from the substrate surface to a selected target substrate presents a major challenge. The utilisation of well-known poly (methyl methacrylate) (PMMA)-assisted graphene transfer promotes defects, impurities, folds, and wrinkles in the graphene sheets, thus affecting its properties. Consequently, the present study demonstrated a polymer-free graphene sheets transfer technique on a Ni substrate derived from cooking palm oil. A dropwise hexane layer substituted the PMMA supporting layer during the etching process to remove the Ni substrate. The quality of the graphene sheet was investigated with optical microscopy by employing a Leica DM1750 M microscope, scanning electron microscopy (SEM) with a Hitachi S-3400N, and Raman spectroscopy utilising a UniDRON automated microscope Raman mapping system with 514 nm laser excitation
Robust sensor fusion for autonomous UAV navigation in GPS denied forest environment
No full textForestry and precision biodiversity data collection using UAVs can be cost-effective and time efficient solution. However, navigating in the forest canopy autonomously can be quite challenging because of its GPS denied environment, cluttered, dynamic, and large scale. Most of the commercial UAVs used in forest applications apply GPS-based navigation which is not suitable for navigating under the canopy. In this paper, an autonomous UAV flight mission in a cluttered forest-like canopy environment is presented. A robust multi-sensor fusion-based robust navigation method which has failure detection features is proposed to enable safe and reliable autonomous navigation. The autonomy architecture utilizes the navigation, planning, and control capabilities of the UAV in a simultaneous manner. To achieve autonomous missions in forest environment, the proposed system has been tested rigorously in a simulated environment and the result shows the capability of autonomous flights in such a challenging environment. The performance of the autonomous flight was evaluated based on mean velocity and path length with respect to the increasing number of trees in the forest
Bioheat transfer of blood flow on healthy and unhealthy bifurcated artery: stenosis
No full textThe development and progression of stenosis with a high probability of rupture can be altered by changing the heat distribution in the bifurcated artery. The purpose of this study is to investigate the behavior of blood flow from healthy to unhealthy artery bifurcation under the influence of bioheat transfer. The blood flow is modelled as laminar, two-dimensional, steady, incompressible, and characterised as a Newtonian fluid. The Galerkin weighted residual (GWR) method is utilised to solve the governing equations. The behaviour of the blood flow due to different Reynolds numbers and the severity of stenosis is graphically investigated and discussed. The results reveal that the velocity profiles increase as the degrees of constriction increase. However, a lower Reynolds number enhances the axial velocity in an artery. The blood flow behaves differently from healthy to unhealthy on the bifurcated artery in the presence of an arterial constriction. The behaviour is significantly affected by the rate of heat transfer and the location of stenosis