The increasing number of diabetes individuals in the globe has alarmed the medical sector to seek alternatives to improve their medical technologies. Machine learning and deep learning approaches are active research in developing intelligent and efficient diabetes detection systems. This study profoundly investigates and discusses the impacts of the latest machine learning and deep learning approaches in diabetes identification/classifications. It is observed that diabetes data are limited in availability. Available databases comprise lab-based and invasive test measurements. Investigating anthropometric measurements and non-invasive tests must be performed to create a cost-effective yet high-performance solution. Several findings showed the possibility of reconstructing the detection models based on anthropometric measurements and non-invasive medical indicators. This study investigated the consequences of oversampling techniques and data dimensionality reduction through feature selection approaches. The future direction is highlighted in the research of feature selection approaches to improve the accuracy and reliability of diabetes identifications.

Gesture recognition technology enables machines to understand human gestures. The technology is considered as a key enabler for gaming and virtual reality applications. In this paper, we propose an effective, low-cost capacitive sensor device to recognize hand gestures. In particular, we designed a prototype of a wearable capacitive sensor unit to capture the capacitance values from the electrodes placed on finger phalanges. The sensor captures finger capacitance values. Each gesture has specific finger capacitance values. We applied a running median filter to the output of the sensor and extracted 15 features for gesture classification training and testing tasks. Subsequently, various analyses were performed to provide more insights into the sensing data. We applied and compared two machine learning algorithms: Error Correction Output Code Support Vector Machines (ECOC-SVM) and K-Nearest Neighbour (KNN) classifiers. The training and testing recognition rates were observed for both intra-participant and inter-participant data sets. Further, we introduced a feature compression approach derived from correlation analysis to reduce the complexity of the machine learning algorithms. Using cross validation, we achieved a classification rate of approximately 99% for intra-participant data. We achieved a lower recognition rate of 97% (average cross validation testing) for compressed feature data set using both machine learning approaches. For the inter-participant data, the recognition rate was 99% (normalized feature data) using KNN and 97% using ECOC-SVM. The research findings show that our recognition system is competitive and has an immense potential for further study.

Early detection of Diabetic Retinopathy (DR) is crucial as it may cause blindness. Manual diagnosis of DR severity by ophthalmologists is challenging and time consuming. Therefore, there has been a significant focus on developing an automated system for identifying DR using retinal fundus images. Recent research has revealed that utilizing pre-trained deep learning networks for diverse image classification tasks provides notable benefits in this context. In this paper, a Transfer Learning (TL) approach with optimized feature weights and parameters is proposed for DR detection and grading tasks. To obtain better generalization during training and to optimize classification, features are extracted from the average pooling layers and fed to an Error Correction Output Code (ECOC) ensemble configuration. Two pre-trained networks (ShuffleNet and ResNet-18) are considered as each pre-trained network offers a different "point of view" of the fundus images, thereby providing more opportunities for accurate "grade-wise" discrimination. A simultaneous feature selection and parameter tuning of the ensemble is applied to further enhance the overall DR detection and grading. Adaptive Differential Evolution (ADE) is chosen for this purpose because it automatically configures the parameters, eliminating the need for manual parameter selection. In this paper, we evaluate two public domain datasets: 1) APTOS and 2) combination of EyePac + Messidor-2. Simulation results show that our proposed method performs better that the conventional deep learning models and are on a par with the existing research work. In particular, the optimal configuration for APTOS 5-class DR grading achieves an accuracy rate of 82%, while for APTOS 2-class grading, it achieves a higher accuracy rate of 96%. Finally, the best configuration for EyePac + Messidor-2 3-class grading results in 75% accuracy.

Malware detection in current times is increasingly important due to the presence of dangerous malicious software (malware) as well as ransomware in digital cyberspace. Conventional approaches such as using malware features (either static or dynamic or hybrid) have been applied for detection. Advances in Deep Learning (DL) has attracted a lot of interests in applications of malware detection. In particular, the file binaries are fed in to the DL neural network for training and testing. Despite the theoretical basis, we find that overfitting has occurred despite applying precaution, such as applying sufficient dropout layers. Limitations can also be attributed to the final classification layers (fully connected shallow network and softmax classification). In this paper, we apply transfer learning using ShuffleNet and DenseNet-201, two models trained on large dataset to recognize daily objects. This is done with the entire layers frozen to prevent overfitting and an optimal Error correction output code (ECOC) ensemble configuration of Support Vector Machines (SVM). Several ECOC coding matrices were applied, e.g., One vs. All (OVA), One vs. One (OVO), Dense Random (DR), and Sparse Random (SR). Each of these configurations represents varying complexity and ensemble size and, hence, a tradeoff between computation reduction and complex non-linear separation appears. Given that the continuous values of SVM parameters may take up high computation for acquiring the optimal parameter configuration, we apply discrete values combination using a grid search approach for parameter optimization. We test the proposed model on Malimg, MaleVis, virus-MNIST, and Dumpware10 datasets. The results show better/comparable accuracy compared with the existing work. The best/average accuracy values for each dataset over 10 trials are: Malimg (99.14%/98.87%), MaleVis (95.01%/93.91%), Virus-MNIST (86.36%/85.79%), Dumpware10 (96.62%/95.79%).

Generalized Frequency Division Multiplexing (GFDM) has been considered as an attractive candidate to replace Orthogonal Frequency Division Multiplexing (OFDM) for the fifth generation (5G) mobile networks. GFDM system has better spectral characteristics compared to the OFDM system due to the usage of properly selected pulse shaping filters. Non-causal ideal filters, such as the raised cosine (RC), are commonly used in the GFDM systems. In practical implementation, non-causal filters need to be truncated and shifted, which will increase the out-of-band (OOB) radiation of the signal and will introduce delay to the system. High OOB radiation will cause interferences between the adjacent channels, and thus should be minimized. This paper proposes to minimize the OOB radiation of the GFDM system using the designed pulse shaping filters. The pulse shaping filters are designed using the computationally efficient quadratic programming (QP) approach. Numerical results illustrated that OOB radiation level of the GFDM system is reduced when the QP filters are used compared to the conventional RC filter. Further investigation shows that the usage of the QP filters are efficient in increasing the high power amplifier (HPA) efficiency, improving the spectral efficiency, and reducing the BER at the receiver.