Accurate animal sound detection and classification may assist in decision-making for many applications such as monitoring the range shift of animals due to climate change, biodiversity assessment of an area, and alerting the nearby people to avoid human-animal conflicts. In this study, a composite deep-learning model for classifying and detecting animal sounds is proposed. The model combines bidirectional long short-term memory (LSTM) and sequential convolutional neural network (CNN). The proposed model consists of four convolution layers with max-pooling, two dropout layers, a bidirectional LSTM, and finally, a fully connected layer. The model is evaluated using a dataset consisting of 885 sound clips of 11 different wild and pet animals. To ensure robust training of the composite deep learning model and make it uniform to the maximum possible animal sound disturbances, the used dataset is augmented for feature extraction. The dataset is augmented by adding noise, stretching, rolling, pitch shifting, etc. Several features are extracted from the augmented dataset, including Chroma, Short-Time Fourier Transform (STFT), Mel Spectrogram, Spectral Contrast, Mel frequency cepstral coefficient (MFCC), and Tonnetz. Extracted features are passed as input for training, and model parameters are optimized for better performance. The performance of the proposed model is compared with conventional CNN and bidirectional LSTM separately using various optimizers. CNN and Bidirectional LSTM achieved the highest accuracy of 91.92% and 93.64%, respectively, with Adam optimizer. Simultaneously, the proposed model achieved the highest accuracy of 95.18 % with the Adagrad optimizer.

Web attacks penetrate the web applications’ security through unauthorized access to sensitive information, disrupting services, and stealing data. Conventionally, rule-based statistical methods distinguish attackers from legitimate users. However, the training through manually extracted weblog features is time-consuming and requires subject expertise. Additionally, the supervised attack classification method needs massive, labeled weblog data, which is expensive and unfeasible. Also, the unsupervised classification techniques have resolved the labeled data insufficiency problem, but their detection performance is unreliable. Recent studies focus on recognizing web attacks through deep neural network-based anomaly detection. Hence, this study proposes an anomaly detection-based Variational LSTM Autoencoder Deviation Network (VLADEN) for recognizing web attacks from weblogs. This work resolves the aforementioned issues by extracting the aberrant information encoded in weblog request data to detect web attacks. VLADEN works in three stages: data preprocessing, anomaly and reference score generation, and classification. The variational LSTM self-encoding-based reference score generation ensures that the anomaly score deviates from the normal data. The proposed model is experimentally validated on three publicly available datasets (CSIS2010, FWAF, and HTTPParams) and evaluated using AUC-ROC and AUC-PR-based evaluation metrics. The results demonstrate the models’ superior performance in detecting attack requests with minimum domain knowledge and labeled data.

Abstract Alzheimer's disorder (AD) causes permanent impairment in the brain's memory of the cellular system, leading to the initiation of dementia. Earlier detection of Alzheimer's disease in the initial stages is challenging for researchers. Deep learning and machine learning‐based techniques can help resolve many issues associated with brain imaging exploration. Brain MR Images (Brain‐MRI) are used to detect Alzheimer's in computable research work. To correctly categorize the stages of Alzheimer's disease, discriminative features need to be extracted from the MR images. Recently, many studies have used deep learning methods for the early detection of this disorder. However, overfitting degrades the deep learning method's performance because the dataset's selection images are smaller and imbalanced. Some studies could not reach more discriminative and effectual attention‐aware features for Alzheimer's stage classification to increase the model performance. In this paper, we develop a novel hierarchical residual attention learning‐inspired multistage conjoined twin network (HRAL‐CTNN) to classify the stages of Alzheimer's. We used augmentation approaches to scale insufficient and imbalanced data. The HRAL‐CTNN is efficiently overcoming the issues of not obtaining efficient attention‐aware and generative features for Alzheimer's stage classification. The proposed model solved the problem of redundant features by extracting attentive discriminant features, and scaling imbalance data by data augmentation, after that training and validation using HRAL‐CTNN. The execution of this proposed work has been performed on the ADNI MRI dataset. This work achieved outstanding accuracy of 99.97 0.01% and F1 score of 99.30 0.02% for Alzheimer's stage classification. This model proposed by our group outperformed the existing related studies in terms of the model's performance score.

Abstract One of the challenging issues in user‐click data of online advertising is the uneven class distribution which biases classification models. Resampling the data is a popular choice for obtaining class balance. However, oversampling results in overfitting, whilst under‐sampling results in information loss. Moreover, enhancing separability between samples, where the classes overlap closer to the decision boundary, is another challenge, which requires a careful pruning of instances towards increasing the separability in data space. Therefore, in this work, a new hybrid data sampling algorithm SMOTEOSS is designed and evaluated, concatenating the synthetic minority oversampling technique (SMOTE) followed by one‐sided selection (OSS) to balance the class distribution. The working of SMOTEOSS is twofold‐ first, it oversamples the under‐represented class distribution using the SMOTE by generating synthetic instances. However, the generation of synthetic instances closer to the decision boundary directly influences the learning model's decision‐making. Utilising OSS, the proposed method then identifies TOMEKLINKS and eliminates the noisy majority instances whilst eliminating the redundant instances. The proposed method's effectiveness is validated on the FDMA 2012 dataset against 10 state‐of‐the‐art sampling methods utilising the gradient tree boosting learning model. To authenticate SMOTEOSS, a fair comparison is made by conducting experiments on other 10 benchmark imbalanced datasets using 10‐fold cross‐validation. Performance is measured using average precision, recall, F1‐score, G‐mean, the area under curve (AUC) and reduction rate. Results showed that the designed hybrid methodology is an efficient alternative to existing sampling methods. The Wilcoxon signed‐rank test is employed to demonstrate significant differences amidst the proposed and conventional sampling algorithms.

Change in users’ preferences over time is a challenging issue in collaborative filtering models. Recent temporal recommender systems (RSs) use latent temporal features or probabilistic latent transition models. The models designed using latent temporal features remain confined to point estimation only, i.e., considering single predicted values (e.g., rating) and completely ignore the likelihood of other possible items being generated due to change in users’ preferences. The latent transitive models represent users’ preference over different states using distribution estimation but ignore user–item features. This article proposes an integrated approach for designing collaborative filtering models using both point and distribution estimation techniques together to address this issue. The matrix factorization-based point estimation strategy is applied to integrate user–item correlations and feature information, whereas the hidden Markov model is used for users’ preference distribution over time. Three benchmark datasets were used to evaluate and compare the proposed model’s performance with related state-of-the-art temporal RS models. The proposed model makes the system more robust in handling dynamic users’ preferences.

Enormous records and data are gathered every day. Organization of this data is a challenging task. Topic modeling provides a way to categorize these documents, where high dimensionality of the corpus affects the result of topic model, making it important to apply feature selection or information retrieval process for dimensionality reduction. The requirement for efficient topic modeling includes the removal of unrelated words that might lead to specious coexistence of the unrelated words. This paper proposes an efficient framework for the generation of better topic coherence, where term frequency-inverse document frequency (TF-IDF) and parsimonious language model (PLM) are used for the information retrieval task. PLM extracts the important information and expels the general words from the corpus, whereas TF-IDF re-estimates the weightage of each word in the corpus. The work carried out in this paper improved the topic coherence measure to provide a better correlation among the actual topic and the topics generated from PLM.