Journal of ICT Research and Applications
Not a member yet
359 research outputs found
Sort by
Anomaly Detection for Human Home Activities Using Pattern Based Sequence Classification
Get PDFIn most countries, the old-age people population continues to rise. Because young adults are busy with their work engagements, they have to let the elderly stay at home alone. This is quite dangerous, as accidents at home may happen anytime without anyone knowing. Although sending elderly relatives to an elderly care center or hiring a caregiver are good solutions, they may not be feasible since it may be too expensive over a long-term period. The behavior patterns of elderly people during daily activities can give hints about their health condition. If an abnormal behavior pattern can be detected in advance, then precautions can be taken at an early stage. Previous studies have suggested machine learning techniques for such anomaly detection but most of the techniques are complicated. In this paper, a simple model for detecting anomaly patterns in human activity sequences using Random forest (RF) and K-nearest neighbor (KNN) classifiers is presented. The model was implemented on a public dataset and it showed that the RF classifier performed better, with an accuracy of 85%, compared to the KNN classifier, which achieved 73%
The Evaluation of DyHATR Performance for Dynamic Heterogeneous Graphs
Get PDFDynamic heterogeneous graphs can represent real-world networks. Predicting links in these graphs is more complicated than in static graphs. Until now, research interest of link prediction has focused on static heterogeneous graphs or dynamically homogeneous graphs. A link prediction technique combining temporal RNN and hierarchical attention has recently emerged, called DyHATR. This method is claimed to be able to work on dynamic heterogeneous graphs by testing them on four publicly available data sets (Twitter, Math-Overflow, Ecomm, and Alibaba). However, after further analysis, it turned out that the four data sets did not meet the criteria of dynamic heterogeneous graphs. In the present work, we evaluated the performance of DyHATR on dynamic heterogeneous graphs. We conducted experiments with DyHATR based on the Yelp data set represented as a dynamic heterogeneous graph consisting of homogeneous subgraphs. The results show that DyHATR can be applied to identify link prediction on dynamic heterogeneous graphs by simultaneously capturing heterogeneous information and evolutionary patterns, and then considering them to carry out link predicition. Compared to the baseline method, the accuracy achieved by DyHATR is competitive, although the results can still be improved
Analytical Approach to Parameter Determination in Kaiser Function for Power-weighted Antenna Array Design
Get PDFWindow methods that are frequently used in the design of finite impulse response filters are also applicable to antenna array designs. This paper explores the application of a Kaiser function in a power-weighted antenna array design, focusing on the determination of the Kaiser function’s β parameter. The determination, which includes the calculation, optimization, and validation of the β parameter, was carried out based on a specific configuration of a linear antenna array design. The observation of this exploration emphasized the suppression of the sidelobe level (SLL) and the width of main lobe (WML) performance. By changing the β parameter, the Kaiser function is capable of approximating different window methods, since it plays an important role in defining the set of weighting coefficients for a specifically targeted SLL. Kaiser function application in power-weighted antenna array designs with a linear arrangement indicates the need of β parameter optimization because of the disagreement between the obtained SLL and the targeted SLL. The optimized β parameter produced a smaller SLL error for even and odd numbers of elements. From the validation, the average SLL error percentage for a targeted SLL of 25 dB, 35 dB, and 45 dB was 6%, 4.31%, 6.10%, respectively
Breast Cancer Diagnosis in Women Using Neural Networks and Deep Learning
Get PDFBreast cancer is a deadly disease affecting women around the world. It can spread rapidly into other parts of the body, causing untimely death when undetected due to rapid growth and division of cells in the breast. Early diagnosis of this disease tends to increase the survival rate of women suffering from the disease. The use of technology to detect breast cancer in women has been explored over the years. A major drawback of most research in this area is low accuracy in the detection rate of breast cancer in women. This is partly due to the availability of few data sets to train classifiers and the lack of efficient algorithms that achieve optimal results. This research aimed to develop a model that uses a machine learning approach (convolution neural network) to detect breast cancer in women with significantly high accuracy. In this paper, a model was developed using 569 mammograms of various breasts diagnosed with benign and maligned cancers. The model achieved an accuracy of 98.25% and sensitivity of 99.5% after 80 iterations.
Energy Consumption Prediction Using Data Reduction and Ensemble Learning Techniques
Get PDFBuilding energy problems have various kinds of aspects, one of which is the difficulty of measuring energy efficiency. With current data development, energy efficiency measurements can be made by developing predictive models to estimate future building needs. However, with the massive amount of data, several problems arise regarding data quality and the lack of scalability in terms of computation memory and time in modeling. In this study, we used data reduction and ensemble learning techniques to overcome these problems. We used numerosity reduction, dimension reduction, and a LightGBM model based on boosting added with a bagging technique, which we compared with incremental learning. Our experimental results showed that the numerosity reduction and dimension reduction techniques could speed up the training process and model prediction without reducing the accuracy. Testing the ensemble learning model also revealed that bagging had the best performance in terms of RMSE and speed, with an RMSE of 262.304 and 1.67 times faster than the model with incremental learning
Early Detection of Stroke for Ensuring Health and Well-Being Based on Categorical Gradient Boosting Machine
Get PDFStroke is believed to be among the leading causes of adult disability worldwide. It is wreaking havoc on African people, families, and governments, with ramifications for the continent’s socio-economic development. On the other hand, stroke research output is insufficient, resulting in a dearth of evidence-based and context-driven guidelines and strategies to combat the region’s expanding stroke burden. Indeed, for African and other developing economies to meet the UN Sustainable Development Goals (SDGs), particularly SDG 3, which aims to guarantee healthy lifestyles and promote well-being for people of all ages, the issue of stroke must be addressed to reduce early death from non-communicable illnesses. This study sought to create a robust predictive model for early stroke diagnosis using an understandable machine learning (ML) technique. We implemented a categorical gradient boosting machine model for early stroke prediction to protect patients’ health and well-being. We compared the effectiveness of our proposed model to existing state-of-the-art machine learning models and previous studies by empirically testing it on a real-world public stroke dataset. The proposed model outperformed the others when compared to the other methods using the research data, achieving the maximum accuracy (96.56%), the area under the curve (AUC) (99.73%), F1-measure (96.68%), recall (99.24%), and precision (93.57%). Functional outcome prediction models based on machine learning for stroke were verified and shown to be adaptable and helpful
Compact and Robust MFCC-based Space-Saving Audio Fingerprint Extraction for Efficient Music Identification on FM Broadcast Monitoring
Get PDFThe Myanmar music industry urgently needs an efficient broadcast monitoring system to solve copyright infringement issues and illegal benefit-sharing between artists and broadcasting stations. In this paper, a broadcast monitoring system is proposed for Myanmar FM radio stations by utilizing space-saving audio fingerprint extraction based on the Mel Frequency Cepstral Coefficient (MFCC). This study focused on reducing the memory requirement for fingerprint storage while preserving the robustness of the audio fingerprints to common distortions such as compression, noise addition, etc. In this system, a three-second audio clip is represented by a 2,712-bit fingerprint block. This significantly reduces the memory requirement when compared to Philips Robust Hashing (PRH), one of the dominant audio fingerprinting methods, where a three-second audio clip is represented by an 8,192-bit fingerprint block. The proposed system is easy to implement and achieves correct and speedy music identification even on noisy and distorted broadcast audio streams. In this research work, we deployed an audio fingerprint database of 7,094 songs and broadcast audio streams of four local FM channels in Myanmar to evaluate the performance of the proposed system. The experimental results showed that the system achieved reliable performance
Mobile Robot Path Planning Optimization Based on Integration of Firefly Algorithm and Cubic Polynomial Equation
Get PDFMobile Robot is an extremely essential technology in the industrial world. Optimal path planning is essential for the navigation of mobile robots. The firefly algorithm is a very promising tool of Swarm Intelligence, which is used in various optimization areas. This study used the firefly algorithm to solve the mobile robot path-planning problem and achieve optimal trajectory planning. The objective of the proposed method is to find the free-collision-free points in the mobile robot environment and then generate the optimal path based on the firefly algorithm. It uses the A∗ algorithm to find the shortest path. The essential function of use the firefly algorithm is applied to specify the optimal control points for the corresponding shortest smooth trajectory of the mobile robot. Cubic Polynomial equation is applied to generate a smooth path from the initial point to the goal point during a specified period. The results of computer simulation demonstrate the efficiency of the firefly algorithm in generating optimal trajectory of mobile robot in a variable degree of mobile robot environment complexity
Wireless Vibration Monitoring System for Milling Process
Get PDFThe implementation of industrial revolution 4.0 in manufacturing industries is necessary to adapt to the rapid changes of technologies. The milling process is one of the common manufacturing processes applied in the industries to produce engineering products. The vibration that occurs in the milling process can disturb the continuity of the process. The wired vibration monitoring system implemented in the manufacturing process needs to be replaced with the wireless monitoring system. Hence wireless vibration monitoring system is developed to solve the problem with wired monitoring systems where tucked cable and high cost are the main challenges of the wired monitoring system. The wireless monitoring system setup is built using three components: sensor node, monitoring node, and base station. Milling experiments with various depths of cut, feed rate, and spindle speed were conducted to examine the performance of the wireless monitoring system. The results indicate the wireless system shows similar data recorded by the wired system. The wireless vibration monitoring system can identify the effect of milling parameters such as depth of cut, feed rate, and spindle speed on the vibrations level. The effect of cut depth is more significant than spindle speed and feed rate in the defined parameters
Strengthening INORMALS Using Context-based Natural Language Generation
Get PDFThe noiseless steganography method that has been proposed by Wibowo can embed up to six characters in the provided cover text, but more than 59% of Indonesian words have a length of more than six characters, so there is room to improve Wibowo’s method. This paper proposes an improvement of Wibowo’s method by modifying the shifting codes and using context-based language generation. Based on 300 test messages, 99% of messages with more than six characters could be embedded by the proposed method, while using Wibowo’s method this was only 34%. Wibowo’s method can embed more than six characters only if the number of shifting codes is less than three, while the proposed method can embed more than six characters even if there are more than three shifting codes. Furthermore, the security for representing the number of code digits is increased by introducing a private key with the probability of guessing less than 1, while in Wibowo’s method this is 1. The naturalness of the cover sentences generated by the proposed method was maintained, which was about 99% when using the proposed method, while it was 98.61% when using Wibowo’s method