From last decade, machine learning (ML) techniques have been used for Urdu text processing. Due to lack of language resources, potential of deep learning (DL) models have not been exploited yet for Urdu text document classification. A text document has more noise, redundant information, and large vocabulary than short text like tweets. This study is the systematic comparison of four well-known DL models. We also compare DL models with four ML models. We also explore the various text preprocessing techniques. Experimental results show that CNN outperforms the others. Further, single-layer architecture of LSTM and BiLSTM performs better than multiple-layers architecture.
Driver Drowsiness Detection (D3) is a challenging task as it requires analysis based on various behavioral and physiological signs such as health issues, mental stress, and exhaustion. Data analytics reveals that driver drowsiness is the reason for one-fifth of all traffic accidents worldwide. Thus, safety devices are valuable for alerting sleepy drivers regarding more danger that may occur. Constant real-time drowsiness detection in complex conditions and denoting is still an open issue. However, facing these challenges, this article proposed a technique called Driver Drowsiness Detection using Custom Deep Learning Model (D3-CDLM). This approach contains four different modules: In the given procedure the investigation and exploration include, (1) feature extraction and selection (2) machine learning and ensemble methods (3) deep learning (4) and the combination of the two, Hybrid. The first operation computes HOG, which stands for Histogram of Oriented Gradient, which is rotation and illumination invariant and resistant to the information in the local areas. Then Principal Component Analysis or PCA is used to obtain the best or top HOG features that are used as inputs to machine learning and ensemble methods-based modules. For hard to learn facial features, transfer learning is also carried out, and a new 30-layer CNNs structure is proposed called CDLM. Finally, the hybrid module’s top features are investigated using the PCA control of the architecture in coordination with the proposed CDLM for detecting drowsiness. Empirical analysis that encompasses all districts were applied on Yawning Detection Dataset. The results reveal that the developed and designed deep learning and hybrid modules acquire better accuracy than the proposed and utilized machine learning-based module along with the compared existing approaches in the pertinent literature.
Pedestrian gender classification is one of the key assignments of pedestrian study, and it finds practical applications in content-based image retrieval, population statistics, human–computer interaction, health care, multimedia retrieval systems, demographic collection, and visual surveillance. In this research work, gender classification was carried out using a deep learning approach. A new 64-layer architecture named 4-BSMAB derived from deep AlexNet is proposed. The proposed model was trained on CIFAR-100 dataset utilizing SoftMax classifier. Then, features were obtained from applied datasets with this pre-trained model. The obtained feature set was optimized with ant colony system (ACS) optimization technique. Various classifiers of SVM and KNN were used to perform gender classification utilizing the optimized feature set. Comprehensive experimentation was performed on gender classification datasets, and proposed model produced better results than the existing methods. The suggested model attained highest accuracy, i.e., 85.4%, and 92% AUC on MIT dataset, and best classification results, i.e., 93% accuracy and 96% AUC, on PKU-Reid dataset. The outcomes of extensive experiments carried out on existing standard pedestrian datasets demonstrate that the proposed framework outperformed existing pedestrian gender classification methods, and acceptable results prove the proposed model as a robust model.