Manual Feature Selection Research Articles

An efficient convolution neural network method for copy-move video forgery detection

Unmanned systems play a pivotal role in military surveillance, critical infrastructure protection, law enforcement, search and rescue operations, and border security, showcasing their multifaceted importance across diverse applications. Video fraud detection is integral to multimedia security, where our task involves the precise identification of modified segments within video sequences. Current approaches to video fraud detection often rely on manual feature selection and models tailored to detect specific tampering types, such as copy-move or splicing. The general representation powers of deep learning models and the connection of multiple forensic characteristics are still not fully explored. This research uses a convolutional neural network (CNN) model to identify copy-move video forgeries. Copy-move forgery is a type of video tampering whereby a portion of the video is copied and pasted somewhere different in the same video to cover an essential video characteristic. The method that is being proposed involves dividing the video into individual frames, extracting features from each frame by using a CNN model that has already been trained, and then utilizing these features to train a new CNN model that would classify each frame as either legitimate or fabricated. The proposed method effectively detects copy-move video forgery with an exceptionally high accuracy rate, exceeding current methods of accuracy and computational effort. The proposed method outperformed all other approaches on the SULFA, GRIP, and VTD datasets. The model's accuracy was 85.42 %, 86.16 %, and 81.87 %, respectively, with the shortest times recorded being 9.6 sec, 11.4 sec, and 13.7 sec, respectively. Consequently, specialists can employ the suggested method as a machine-learning instrument for detecting fake videos in real-time.

Heartbeat classification using various machine learning models: A comparative study

Cardiac arrhythmias, known as irregular heartbeats, pose a notable health threat that necessitates prompt diagnosis, as untreated arrhythmias can lead to severe heart complications. Among the various methods for arrhythmia detection, electrocardiography is the most prevalent due to its non-invasive monitoring of heart activity. However, manual electrocardiogram (ECG) analysis is inefficient and prone to errors, prompting the exploration of machine learning (ML) models for ECG feature recognition. Integrating ML models with ECG analysis can revolutionize cardiac diagnostics by improving healthcare efficiency and outcomes by enhancing the accuracy and consistency of existing approaches as well as their processing speed for large datasets. Unfortunately, current ML methods encounter two key limitations: prolonged training times and the need for manual feature selection. To address these issues, we propose using ML models enhanced with innovative techniques such as the Fourier transform (FT) and Gaussian noise injection for improved cardiac health assessment. To validate this approach, we utilized statistical tools, including Pearson correlation and p-values, to uncover relationships within the data. In addition, we employed the FT technique to extract and analyze frequency-domain features. Our comparative study of different ML models relied on metrics such as accuracy, precision, recall, F1 score, and receiver operating characteristic area under the receiver operating characteristic curve, demonstrating XGBoost’s impressive average recall of 0.956 with 99.96% overall accuracy. An average precision of 0.956 further underscored the accuracy of XGBoost’s predictions, indicating its high level of reliability in distinguishing various cardiac conditions. These results highlight the considerable potential of ML techniques for precise ECG-based clinical diagnoses, helping healthcare professionals make more accurate and timely decisions in patient care.

DynProfiler: a Python package for comprehensive analysis and interpretation of signaling dynamics leveraged by deep learning techniques.

Signaling dynamics encode important features and regulatory mechanisms of biological systems, and recent studies have reported the use of simulated signaling dynamics with mechanistic modeling as biomarkers for human diseases. Given the success of deep learning techniques, it is expected that they can extract informative patterns from simulation results more effectively than traditional approaches involving manual feature selection, which can be used for subsequent analyses, such as patient stratification and survival prediction. Here, we propose DynProfiler, which utilizes the entire signaling dynamics, including intermediate variables, as input and leverages deep learning techniques to extract informative features without requiring any labels. Furthermore, DynProfiler incorporates a modern explainable AI solution to provide quantitative time-dependent importance scores for each dynamics. Using simulated dynamics of patients with breast cancer as an example, we demonstrate DynProfiler's ability to extract high-quality features that can predict mortality risk and identify important dynamics, highlighting upregulated phosphorylated GSK3β as a biomarker for poor prognosis. Overall, this tool can be useful for clinical application, as well as for elucidating biological system dynamics. The DynProfiler Python library is available in GitHub at https://github.com/okadalabipr/DynProfiler.

FPGA‐Based Implementation of Real‐Time Cardiologist‐Level Arrhythmia Detection and Classification in Electrocardiograms Using Novel Deep Learning

ABSTRACTCardiac arrhythmia refers to irregular heartbeats caused by anomalies in electrical transmission in the heart muscle, and it is an important threat to cardiovascular health. Conventional monitoring and diagnosis still depend on the laborious visual examination of electrocardiogram (ECG) devices, even though ECG signals are dynamic and complex. This paper discusses the need for an automated system to assist clinicians in efficiently recognizing arrhythmias. The existing machine‐learning (ML) algorithms have extensive training cycles and require manual feature selection; to eliminate this, we present a novel deep learning (DL) architecture. Our research introduces a novel approach to ECG classification by combining the vision transformer (ViT) and the capsule network (CapsNet) into a hybrid model named ViT‐Cap. We conduct necessary preprocessing operations, including noise removal and signal‐to‐image conversion using short‐time Fourier transform (SIFT) and continuous wavelet transform (CWT) algorithms, on both normal and abnormal ECG data obtained from the MIT‐BIH database. The proposed model intelligently focuses on crucial features by leveraging global and local attention to explore spectrogram and scalogram image data. Initially, the model divides the images into smaller patches and linearly embeds each patch. Features are then extracted using a transformer encoder, followed by classification using the capsule module with feature vectors from the ViT module. Comparisons with existing conventional models show that our proposed model outperforms the original ViT and CapsNet in terms of classification accuracy for both binary and multi‐class ECG classification. The experimental findings demonstrate an accuracy of 99% on both scalogram and spectrogram images. Comparative analysis with state‐of‐the‐art methodologies confirms the superiority of our framework. Additionally, we configure a field‐programmable gate array (FPGA) to implement the proposed model for real‐time arrhythmia classification, aiming to enhance user‐friendliness and speed. Despite numerous suggestions for high‐performance FPGA accelerators in the literature, our FPGA‐based accelerator utilizes optimization of loop parallelization, FP data, and multiply accumulation (MAC) unit. Our accelerator architecture achieves a 57% reduction in processing time and utilizes fewer resources compared to a floating‐point (FlP) design.

Performance Analysis of Deep Learning based Signal Constellation Identification Algorithms for Underwater Acoustic Communications

This research delves into the evaluation of Deep learning signal constellation identification (DL-SCI) algorithms in underwater acoustic communications using Orthogonal Frequency Division Multiplexing (OFDM). It distinctly examines at how effective the recurrent neural networks (RNNs), particularly, Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) algorithms in predicting the signal constellation when applied to different underwater acoustic channels characteristics. Unlike manual feature selection in machine learning (ML), in this paper, DL-SCI exploits the labelled OFDM signals at the transmitter to detect and decode them at the receiver. In order to measure their effectiveness performance metrics, Bit Error Rate (BER) and parameters derived from the confusion matrix such as accuracy and precision are used. The study highlights the importance of utilizing zero cyclic prefix techniques which can exploit the inherent bandwidth limitation effectively. Furthermore, when examining complexity, it is observed that both GRU and LSTM algorithms require less floating-point operations (FLOPS) compared to traditional methods such as Minimum Mean Square Error (MMSE) and Least Squares (LS). Interestingly GRU shows performance in terms of complexity when compared to LSTM. Moreover, GRU outperforms LSTM by achieving a 4 dB improvement for long subcarriers. These results emphasize the effectiveness of learning techniques in enhancing performance and efficiency in acoustic communications.

Deep Learning-Based Defects Detection in Keyhole TIG Welding with Enhanced Vision.

Keyhole tungsten inert gas (keyhole TIG) welding is renowned for its advanced efficiency, necessitating a real-time defect detection method that integrates deep learning and enhanced vision techniques. This study employs a multi-layer deep neural network trained on an extensive welding image dataset. Neural networks can capture complex nonlinear relationships through multi-layer transformations without manual feature selection. Conversely, the nonlinear modeling ability of support vector machines (SVM) is limited by manually selected kernel functions and parameters, resulting in poor performance for recognizing burn-through and good welds images. SVMs handle only lower-level features such as porosity and excel only in detecting simple edges and shapes. However, neural networks excel in processing deep feature maps of "molten pools" and can encode deep defects that are often confused in keyhole TIG. Applying a four-class classification task to weld pool images, the neural network adeptly distinguishes various weld states, including good welds, burn-through, partial penetration, and undercut. Experimental results demonstrate high accuracy and real-time performance. A comprehensive dataset, prepared through meticulous preprocessing and augmentation, ensures reliable results. This method provides an effective solution for quality control and defect prevention in keyhole TIG welding process.

Classification of osteoarthritic and healthy cartilage using deep learning with Raman spectra

Raman spectroscopy is a rapid method for analysing the molecular composition of biological material. However, noise contamination in the spectral data necessitates careful pre-processing prior to analysis. Here we propose an end-to-end Convolutional Neural Network to automatically learn an optimal combination of pre-processing strategies, for the classification of Raman spectra of superficial and deep layers of cartilage harvested from 45 Osteoarthritis and 19 Osteoporosis (Healthy controls) patients. Using 6-fold cross-validation, the Multi-Convolutional Neural Network achieves comparable or improved classification accuracy against the best-performing Convolutional Neural Network applied to either the raw or pre-processed spectra. We utilised Integrated Gradients to identify the contributing features (Raman signatures) in the network decision process, showing they are biologically relevant. Using these features, we compared Artificial Neural Networks, Decision Trees and Support Vector Machines for the feature selection task. Results show that training on fewer than 3 and 300 features, respectively, for the disease classification and layer assignment task provide performance comparable to the best-performing CNN-based network applied to the full dataset. Our approach, incorporating multi-channel input and Integrated Gradients, can potentially facilitate the clinical translation of Raman spectroscopy-based diagnosis without the need for laborious manual pre-processing and feature selection.

Mental fatigue recognition study based on 1D convolutional neural network and short-term ECG signals.

Mental fatigue has become a non-negligible health problem in modern life, as well as one of the important causes of social transportation, production and life accidents. Fatigue detection based on traditional machine learning requires manual and tedious feature extraction and feature selection engineering, which is inefficient, poor in real-time, and the recognition accuracy needs to be improved. In order to recognize daily mental fatigue level more accurately and in real time, this paper proposes a mental fatigue recognition model based on 1D Convolutional Neural Network (1D-CNN), which inputs 1D raw ECG sequences of 5s duration into the model, and can directly output the predicted fatigue level labels. The fatigue dataset was constructed by collecting the ECG signals of 22 subjects at three time periods: 9:00-11:00 a.m., 14:00-16:00 p.m., and 19:00-21:00 p.m., and then inputted into the 19-layer 1D-CNN model constructed in the present study for the classification of mental fatigue in three grades. The results showed that the model was able to recognize the fatigue levels effectively, and its accuracy, precision, recall, and F1 score reached 98.44%, 98.47%, 98.41%, and 98.44%, respectively. This study further improves the accuracy and real-time performance of recognizing multi-level mental fatigue based on electrocardiography, and provides theoretical support for real-time fatigue monitoring in daily life.

High-Resolution Downscaling with Interpretable Relevant Vector Machine: Rainfall Prediction for Case Study in Selangor

Due to the discrepancy in resolution between existing global climate model output and the resolution required by decision-makers, there is a persistent need for climate downscaling. We conducted a study to determine the effectiveness of Relevant Vector Machine (RVM), one of the machine learning approaches, in outperforming existing statistical methods in downscaling historical rainfall data in the complex terrain of Selangor, Malaysia. While machine learning eliminates the requirement for manual feature selection when extracting significant information from predictor fields, considering multiple pivotal factors is essential. These factors include identifying relevant atmospheric features contributing to rainfall, addressing missing data, and developing a significant model to predict daily rainfall intensity using appropriate machine-learning techniques. The Principal Component Analysis (PCA) technique was employed to choose relevant environmental variables as input for the machine learning model, and various imputation methods were utilized to manage missing data, such as mean imputation and the KNN algorithm. To assess the performance of the RVM-based rainfall model, we collected a dataset from the Department of Irrigation and Drainage Malaysia. We used Nash-Sutcliffe Efficiency (NSE) and Root Mean Square Error (RMSE) as evaluation metrics. This study concluded that Relevance Vector Machine (RVM) models are suitable for forecasting future rainfall since they can support large rainfall extremes and generate reliable daily rainfall estimates based on rainfall extremes. In this study, the RVM model was employed to determine a predictive association between predictand variables and predictors.

Sydney’s residential relocation landscape: Machine learning and feature selection methods unpack the whys and whens

This study investigates household residential relocation timing, an aspect vital for transport and urban planning. Analyzing a high-dimensional dataset from 1,024 relocations in Sydney, Australia, the research contrasts ten machine learning survival techniques with three classical survival models. Results indicate that when classical models are paired with tree-based automated feature selectors, they align closely with machine learning outcomes. Notably, the GBM, XGBoost, and Random Forest models emerge as standout performers. The study provides a comprehensive comparison between automatic and manual feature selection, shedding light on variables influencing households’ duration of stay. While stacked ensemble modeling, which leverages predictions from various models, is used to enhance accuracy, the improvements are marginal, underscoring inherent modeling challenges, particularly the recurring issue of misclassifying specific pairs of households in the concordance index measure. A thorough feature analysis highlights homeownership as the foremost predictor, underscoring the importance of recent life events and accessibility features in relocation decisions. The research emphasizes the importance of considering the accessibility of both current and future homes in relocation models, with 20% feature significance in model outcomes. Building on these foundational insights, the study paves the way for a deeper understanding of individual decision-making processes in sustainable urban planning.

A prediction model of nonclassical secreted protein based on deep learning

AbstractMost of the current nonclassical proteins prediction methods involve manual feature selection, such as constructing features of samples based on the physicochemical properties of proteins and position‐specific scoring matrix (PSSM). However, these tasks require researchers to perform some tedious search work to obtain the physicochemical properties of proteins. This paper proposes an end‐to‐end nonclassical secreted protein prediction model based on deep learning, named DeepNCSPP, which employs the protein sequence information and sequence statistics information as input to predict whether it is a nonclassical secreted protein. The protein sequence information and sequence statistics information are extracted using bidirectional long‐ and short‐term memory and convolutional neural networks, respectively. Among the experiments conducted on the independent test dataset, DeepNCSPP achieved excellent results with an accuracy of 88.24%, Matthews coefficient (MCC) of 77.01%, and F1‐score of 87.50%. Independent test dataset testing and 10‐fold cross‐validation show that DeepNCSPP achieves competitive performance with state‐of‐the‐art methods and can be used as a reliable nonclassical secreted protein prediction model. A web server has been constructed for the convenience of researchers. The web link is https://www.deepncspp.top/. The source code of DeepNCSPP has been hosted on GitHub and is available online (https://github.com/xiaoliu166370/DEEPNCSPP).

Unsupervised Layer-Wise Score Aggregation for Textual OOD Detection

Out-of-distribution (OOD) detection is a rapidly growing field due to new robustness and security requirements driven by an increased number of AI-based systems. Existing OOD textual detectors often rely on anomaly scores (\textit{e.g.}, Mahalanobis distance) computed on the embedding output of the last layer of the encoder. In this work, we observe that OOD detection performance varies greatly depending on the task and layer output. More importantly, we show that the usual choice (the last layer) is rarely the best one for OOD detection and that far better results can be achieved, provided that an oracle selects the best layer. We propose a data-driven, unsupervised method to leverage this observation to combine layer-wise anomaly scores. In addition, we extend classical textual OOD benchmarks by including classification tasks with a more significant number of classes (up to 150), which reflects more realistic settings. On this augmented benchmark, we show that the proposed post-aggregation methods achieve robust and consistent results comparable to using the best layer according to an oracle while removing manual feature selection altogether.

Detection of Coronary Artery Disease Based on Clinical Phonocardiogram and Multiscale Attention Convolutional Compression Network.

Heart sound is an important physiological signal that contains rich pathological information related with coronary stenosis. Thus, some machine learning methods are developed to detect coronary artery disease (CAD) based on phonocardiogram (PCG). However, current methods lack sufficient clinical dataset and fail to achieve efficient feature utilization. Besides, the methods require complex processing steps including empirical feature extraction and classifier design. To achieve efficient CAD detection, we propose the multiscale attention convolutional compression network (MACCN) based on clinical PCG dataset. Firstly, PCG dataset including 102 CAD subjects and 82 non-CAD subjects was established. Then, a multiscale convolution structure was developed to catch comprehensive heart sound features and a channel attention module was developed to enhance key features in multiscale attention convolutional block (MACB). Finally, a separate downsampling block was proposed to reduce feature losses. MACCN combining the blocks can automatically extract features without empirical and manual feature selection. It obtains good classification results with accuracy 93.43%, sensitivity 93.44%, precision 93.48%, and F1 score 93.42%. The study implies that MACCN performs effective PCG feature mining aiming for CAD detection. Further, it integrates feature extraction and classification and provides a simplified PCG processing case.

Physiological records-based situation awareness evaluation under aviation context: A comparative analysis

Situational Awareness (SA) assessment is of paramount importance in various domains, with particular significance in the military for safe aviation decision-making. It involves encompassing perception, comprehension, and projection levels in human beings. Accurate evaluation of SA statuses across these three levels is crucial for mitigating human false-positive and false-negative rates in monitoring complex scenarios in the aviation context. This study proposes a comprehensive comparative analysis by involving two types of physiological records: electroencephalogram (EEG) signals and brain electrical activity mapping (BEAM) images. These two modalities are leveraged to automate precise SA evaluation using both conventional machine learning and advanced deep learning techniques. Benchmarking experiments reveal that the BEAM-based deep learning models attain state-of-the-art performance scores of 0.955 for both SA perception and comprehension levels, respectively. Conversely, the EEG signals-based manual feature extraction, selection, and classification approach achieved a superior accuracy of 0.929 for the projection level of SA. These findings collectively highlight the potential of deploying diverse physiological records as valuable computational tools for enhancing SA evaluation throughout aviation decision-making safety.

LiteTransNet: An interpretable approach for landslide displacement prediction using transformer model with attention mechanism

Accurate landslide displacement prediction is crucial for effective early warning systems to mitigate hazards. The importance of historical information varies with time during prediction due to the underlying landslide deformation mechanism. Despite advances in dynamic machine learning models like Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), they struggle to fully capture and interpret the varying importance of historical information during prediction, resulting in decreased accuracy and limited understanding of landslide physical mechanisms. Additionally, these approaches rely on manual feature selection, which is independent of the learning process and therefore is prone to estimation errors. To address these challenges, we propose LiteTransNet, a Lightweighted Transformer Network tailored for landslide displacement prediction. Built on the revolutionary Transformer model for sequential data, LiteTransNet leverages localized self-attention to selectively focus on relevant timestamps and provides interpretable results through its attention heatmap. Furthermore, LiteTransNet performs end-to-end time series modeling without extensive feature engineering. We validate LiteTransNet on two landslides in China's Three Gorges Reservoir Area and demonstrate its improved accuracy over recurrent baselines utilizing feature selection methods. Notably, the attention heatmap generated by LiteTransNet provides interpretable insights into the model's temporal dependency learning. It uncovers the varying importance of historical information at different timestamps, showcasing how LiteTransNet's attention aligns with specific periods of the external environment that cause significant disruptions in the landslide system. Our findings reveal that these disruptions continuously impact displacement prediction until their effects fade or are replaced by subsequent intense disturbances, which traditional recurrent methods are unable to capture. Further experiment shows that LiteTransNet enhances efficiency through its streamlined architecture and its parallelization capability. Overall, LiteTransNet innovates landslide prediction by providing accuracy, interpretability, and efficiency, enhancing understanding of deformation mechanisms to facilitate effective early warning systems.

Durian Species Classification Using Deep Learning Method

Durian is a popular fruit in Southeast Asia, and the market offers various species of durians. Accurate species classification is crucial for quality control, grading, and marketing. However, the complexity of this task has led to the utilization of machine learning and deep learning methods. Traditional machine learning algorithms, such as K-Nearest Neighbors, Linear Discriminant Analysis, Support Vector Machines, and Random Forests, have demonstrated good accuracy, but they require extensive feature engineering. Deep learning algorithms, particularly Convolutional Neural Networks, can automatically extract features, making them less dependent on manual feature selection. This research aims to review deep learning classification algorithms, including Convolutional Neural Networks and Recurrent Neural Networks, to determine the most suitable algorithm for an efficient and accurate durian classification system. The objective is to enhance the precision and speed of durian species classification, presenting potential advantages for both durian producers and consumers. The literature review revealed that Convolutional Neural Networks outperformed other deep learning and traditional machine learning algorithms on datasets of varying sizes, achieving the highest accuracy of 98.96% through techniques like image resizing, color conversion, and additional parameters such as days harvested and dry weight. Deep learning emerges as a promising approach for robust and accurate durian species recognition, with future directions including developing models to classify durian species from different plant parts and even real-time video analysis. However, while Convolutional Neural Networks lead the way, a critical research gap exists in identifying optimal features, necessitating further investigation to refine durian species recognition accuracy.

An innovative network intrusion detection system (NIDS): Hierarchical deep learning model based on Unsw-Nb15 dataset

With the increasing prevalence of network intrusions, the development of effective network intrusion detection systems (NIDS) has become crucial. In this study, we propose a novel NIDS approach that combines the power of long short-term memory (LSTM) and attention mechanisms to analyze the spatial and temporal features of network traffic data. We utilize the benchmark UNSW-NB15 dataset, which exhibits a diverse distribution of patterns, including a significant disparity in the size of the training and testing sets. Unlike traditional machine learning techniques like support vector machines (SVM) and k-nearest neighbors (KNN) that often struggle with limited feature sets and lower accuracy, our proposed model overcomes these limitations. Notably, existing models applied to this dataset typically require manual feature selection and extraction, which can be time-consuming and less precise. In contrast, our model achieves superior results in binary classification by leveraging the advantages of LSTM and attention mechanisms. Through extensive experiments and evaluations with state-of-the-art ML/DL models, we demonstrate the effectiveness and superiority of our proposed approach. Our findings highlight the potential of combining LSTM and attention mechanisms for enhanced network intrusion detection.

Synergistic advantages of deep learning and reinforcement learning in economic forecasting

With the progress of science and technology, emerging technologies such as deep learning and reinforcement learning have emerged in economic forecasting, injecting synergy into this field. First of all, deep learning makes the economic model capture the dynamic changes of the economic system more comprehensively and accurately by dealing with nonlinear relations, learning complex characteristics and conducting accurate time series analysis. Its multi-level neural network structure and the ability to learn features automatically improve the adaptability of the model and avoid the tedious process of traditional manual feature selection. Secondly, the introduction of reinforcement learning gives the economic forecasting model more flexible and adaptive advantages. Through interactive learning between agent and environment, the model can optimize decision-making, deal with uncertainty better, and is adaptive. Reinforcement learning improves the flexibility of decision-making by constantly trying and learning and adjusting strategies to adapt to changes in the economic environment. The collaborative application of them has made remarkable progress in nonlinear relationship modeling, feature learning, time series analysis, decision making, uncertainty handling and self-adaptability, which has improved the accuracy and adaptability of the model and enhanced its robustness in the complex and changeable economic environment. In the future, deepening the research and optimization of these emerging technologies in practical application will provide better performance for economic forecasting, provide more comprehensive, accurate and practical information for decision makers, and help more scientific, flexible and fine economic management and policy formulation.

Markerless vision-based functional movement screening movements evaluation with deep neural networks

The functional movement screen (FMS) test is a seven-test battery used to assess fundamental movement abilities of individuals. It is commonly used to predict sports injuries but relies on clinical expertise and is not suitable for self-examination. This study presents an automatic FMS movement assessment framework using a multi-view deep neural network called MVDNN. The framework combines automatic skeleton extraction with manual feature selection to extract 3D trajectory features of human skeleton joints from two different directions. Three mainstream methods of time-series modeling are then used to learn high-level feature representation from skeleton sequences, and motion features from two views are fused to provide complementary information. Results of public FMS movements dataset demonstrate that our MVDNN outperforms current state-of-the-art methods with an average miF1 score of 0.857, maF1 score of 0.768, and Kappa score of 0.640 over ten runs.

Hybrid Deep Learning Techniques for Securing Bioluminescent Interfaces in Internet of Bio Nano Things.

The Internet of bio-nano things (IoBNT) is an emerging paradigm employing nanoscale (~1-100 nm) biological transceivers to collect in vivo signaling information from the human body and communicate it to healthcare providers over the Internet. Bio-nano-things (BNT) offer external actuation of in-body molecular communication (MC) for targeted drug delivery to otherwise inaccessible parts of the human tissue. BNTs are inter-connected using chemical diffusion channels, forming an in vivo bio-nano network, connected to an external ex vivo environment such as the Internet using bio-cyber interfaces. Bio-luminescent bio-cyber interfacing (BBI) has proven to be promising in realizing IoBNT systems due to their non-obtrusive and low-cost implementation. BBI security, however, is a key concern during practical implementation since Internet connectivity exposes the interfaces to external threat vectors, and accurate classification of anomalous BBI traffic patterns is required to offer mitigation. However, parameter complexity and underlying intricate correlations among BBI traffic characteristics limit the use of existing machine-learning (ML) based anomaly detection methods typically requiring hand-crafted feature designing. To this end, the present work investigates the employment of deep learning (DL) algorithms allowing dynamic and scalable feature engineering to discriminate between normal and anomalous BBI traffic. During extensive validation using singular and multi-dimensional models on the generated dataset, our hybrid convolutional and recurrent ensemble (CNN + LSTM) reported an accuracy of approximately ~93.51% over other deep and shallow structures. Furthermore, employing a hybrid DL network allowed automated extraction of normal as well as temporal features in BBI data, eliminating manual selection and crafting of input features for accurate prediction. Finally, we recommend deployment primitives of the extracted optimal classifier in conventional intrusion detection systems as well as evolving non-Von Neumann architectures for real-time anomaly detection.