Feature Extraction Techniques Research Articles

The impact of feature extraction techniques on the performance of text data classification models

<p class="MsoNormal" style="text-align: justify;">Sentiment analysis is a crucial discipline that focuses on the interpretation of feelings and points of view in textual data. Our study aims to assess the impact of different feature extraction methods on the accuracy of opinion research models. Techniques such as bag-of-words (BoW), term frequency-inverse document frequency (TF-IDF), Word2Vec, global vectors (GloVe) and bidirectional encoder representations from transformers (BERT) were used with three machine learning algorithms and three deep learning networks as classifiers. The IMDB movie review dataset was used for evaluation. The results showed that combining BERT with LSTM, CNN and RNN improved performance, achieving an accuracy rate of 94%, precision of 94.14%, recall of 93.27% and an F1 score of 89.33%. These results highlight the significant contribution of ERTB to model performance, outperforming other feature extraction techniques in text classification. The study concludes that the fusion of BERT and LSTM significantly improves model accuracy for opinion retrieval, recommending BERT as the main feature extraction method for optimizing performance in NLP tasks.</p>

FAULT DETECTION AND CLASSIFICATION USING DENSENET, SURF-BASED ANN, AND INFRARED THERMOGRAPHY

The reliability and efficiency of electric motors are critical in industrial applications, where unexpected faults can lead to costly downtime and safety hazards. Traditional fault detection methods often require extensive manual inspection and may not capture subtle anomalies in motor behavior. Infrared thermography has emerged as a non-invasive technique to detect temperature variations in motor components, which can indicate potential faults. However, the challenge lies in accurately classifying these faults to prevent failures. Current methods lack the precision needed to classify various motor faults accurately and quickly, especially when dealing with complex thermal patterns. The integration of advanced deep learning architectures with feature extraction techniques presents an opportunity to enhance the detection and classification of motor faults. This study proposes a hybrid model combining DenseNet, a deep learning architecture known for its high performance in image analysis, with a Speeded-Up Robust Features (SURF)-based Artificial Neural Network (ANN) for feature extraction and classification. Infrared thermography images of motors were first processed through DenseNet for initial feature extraction. The SURF algorithm further refined these features, which were then classified using ANN. The model was trained and validated on a dataset of infrared thermography images, representing various motor fault conditions, including bearing wear, misalignment, and insulation failure. The proposed model achieved an overall accuracy of 98.7% in detecting and classifying motor faults, outperforming traditional methods by 5.2%. The model also demonstrated high sensitivity (97.8%) and specificity (99.1%) in identifying subtle temperature variations indicative of early-stage faults. These results highlight the effectiveness of the DenseNet and SURF-based ANN approach in enhancing the reliability of motor fault detection using infrared thermography.

Optimizing breast cancer diagnosis: combining hybrid architectures through Apache Spark

Early detection and diagnosis of breast cancer are critical for saving lives. This paper addresses two major challenges associated with this task: the vast amount of data processing involved and the need for early detection of breast cancer. To tackle these issues, we developed thirty hybrid architectures by combining five deep learning techniques (Xception, Inception-V3, ResNet50, VGG16, VGG19) as feature extractors and six classifiers (random forest, logistic regression, naive Bayes, gradient-boosted tree, decision tree, and support vector machine) implemented on the Spark framework. We evaluated the performance of these architectures using four classification criteria. The results, analyzed using Scott Knott's statistical test, demonstrated the effectiveness of merging deep learning feature extraction techniques with traditional classifiers for classifying breast cancer into malignant and benign tumors. Notably, the hybrid architecture using logistic regression as the classifier and ResNet50 for feature extraction (RESLR) emerged as the top performer. It achieved impressive accuracy scores of 98.20%, 96.59%, 96.64%, and 94.84% across the Break-His dataset at different magnifications (40X, 100X, 200X, and 400X) respectively. Additionally, RESLR achieved an accuracy of 97.05% on the ICIAR dataset and a remarkable accuracy of 95.31% on the FNAC dataset.

LINEAR REGRESSION-BASED ANALYSIS OF MULTIMEDIA DATA FOR AI-DRIVEN PATTERN RECOGNITION

Multimedia data, encompassing images, videos, and audio, has become a cornerstone in various AI-driven applications, particularly in pattern recognition tasks. The increasing complexity and volume of multimedia data necessitate robust and scalable analytical approaches. Traditional pattern recognition techniques often struggle to effectively manage the high-dimensional and multimodal nature of multimedia data. This study addresses the challenge by leveraging linear regression for analyzing multimedia data to enhance AI-driven pattern recognition. The proposed method integrates linear regression models with feature extraction techniques to identify and map underlying patterns within the multimedia data. The process begins with preprocessing steps, including normalization and dimensionality reduction, to ensure data consistency and manage computational complexity. Subsequently, linear regression models are applied to establish relationships between the extracted features and predefined classes or labels. The model’s performance is evaluated using precision, recall, and F1-score metrics. Experimental results on a benchmark multimedia dataset reveal that the proposed approach achieves an average accuracy of 92.4%, with a precision of 91.8% and a recall of 93.1%. These results outperform several state-of-the-art methods, demonstrating the model’s efficacy in accurately recognizing patterns within diverse multimedia data. Furthermore, the model exhibits scalability, maintaining high performance even when applied to large-scale datasets, thus validating its potential for real-world AI applications. The study concludes that linear regression, when integrated with appropriate feature extraction and preprocessing techniques, offers a viable solution for enhancing AI-driven pattern recognition in multimedia data.

Intelligent diagnosis of magnetorheological clutch using image processing techniques

The clutch is one of the important components of a vehicle system. Therefore, it is crucial to keep an eye on them and spot any faults while they are operating so that the operators can be informed and possible issues can be solved before they arise. In general, a fault in the clutch causes it to get eccentricity in the disc and the contact between housing and disc during its operation which impacts the non-linear dynamic behaviour of the clutch mechanism and hence it affects the performance of the same. For condition monitoring, infrared imaging is essential since the thermographic patterns change based on the defect or machine condition. In this study, novel attempts have been made to present an intelligent tribological diagnosis of magnetorheological clutch under conditions such as eccentricity of the disc due to misalignment and the contact between housing and disc with the help of an image obtained from the thermal camera. The study for intelligent tribological diagnosis has been done by image processing algorithms in the MATLAB environment. The image segmentation for diagnosis is done by the thresholding technique followed by the feature extraction technique by the Circular Hough transform algorithm. The results successfully determine whether the eccentricity and the contact between the housing and disc due to misalignment are present or not, with the help of circular features in red and blue colour.

Enhancing Intercultural Business English Communication Factors Evaluation System Using the Termite Life Cycle Optimization Algorithm and Dynamically Stabilized Recurrent Neural Network

In today's globalized business environment, effective intercultural communication in English is paramount for successful collaboration among professionals from diverse backgrounds. To enhance the accuracy of the evaluation system, enhancing intercultural business English communication factors evaluation system using the termite life cycle optimization algorithm and dynamically stabilized recurrent neural network (IBEC–DSRNN–TLCOA) is proposed in this manuscript. The input image is captured from mobile camera. Then the input images are preprocessed using intelligent weight decreasing firefly–particle filtering (IWDFPF) to remove noise and enhance the input images. Afterwards, the preprocessed image is fed to the entropy-founded spatial fuzzy C-means clustering approach for segmenting the image. Then the contrast, correlation, energy and homogeneousness features are extracted by using force-invariant improved feature extraction technique. The extracted features are given to dynamically stabilized recurrent neural network (DSRNN) to image target detection and English description generation. Termite life cycle optimization algorithm (TLCOA) is employed to enhance the weight parameters of DSRNN. The proposed IBEC–DSRNN–TLCOA method is implemented. The proposed IBEC–DSRNN–TLCOA method provides 32.53%, 31.86%, and 35.72% higher accuracy; 35.58%, 32.16%, and 37.72% higher F-measure when compared with the existing methods, such as exploration of intelligent translation with evaluation systems for business English (IBEC–RCNN), E-learning engagement with convolution neural networks on business education (IBEC–CNN), and deep neural network-based research on scoring business English oral training (IBEC–DNN), respectively.

Machine Learning Meets Meta-Heuristics: Bald Eagle Search Optimization and Red Deer Optimization for Feature Selection in Type II Diabetes Diagnosis.

This article investigates the effectiveness of feature extraction and selection techniques in enhancing the performance of classifier accuracy in Type II Diabetes Mellitus (DM) detection using microarray gene data. To address the inherent high dimensionality of the data, three feature extraction (FE) methods are used, namely Short-Time Fourier Transform (STFT), Ridge Regression (RR), and Pearson's Correlation Coefficient (PCC). To further refine the data, meta-heuristic algorithms like Bald Eagle Search Optimization (BESO) and Red Deer Optimization (RDO) are utilized for feature selection. The performance of seven classification techniques, Non-Linear Regression-NLR, Linear Regression-LR, Gaussian Mixture Models-GMMs, Expectation Maximization-EM, Logistic Regression-LoR, Softmax Discriminant Classifier-SDC, and Support Vector Machine with Radial Basis Function kernel-SVM-RBF, are evaluated with and without feature selection. The analysis reveals that the combination of PCC with SVM-RBF achieved a promising accuracy of 92.85% even without feature selection. Notably, employing BESO with PCC and SVM-RBF maintained this high accuracy. However, the highest overall accuracy of 97.14% was achieved when RDO was used for feature selection alongside PCC and SVM-RBF. These findings highlight the potential of feature extraction and selection techniques, particularly RDO with PCC, in improving the accuracy of DM detection using microarray gene data.

Acoustic-based diagnostics for UAV propeller damage using HNR and Gaussian Naive Bayes

PurposeThe purpose of this study is to determine propeller damage based on acoustic recordings taken from unmanned aerial vehicle (UAV) propellers operated at different thrust conditions on a test bench. Propeller damage is especially critical for fixed-wing UAVs to sustain a safe flight. The acoustic characteristics of the propeller vary with different propeller damages.Design/methodology/approachFor the research, feature extraction methods and machine learning techniques were used during damage detection from propeller acoustic data. First of all, sound recordings were obtained by operating five different damaged propellers and undamaged propellers under three different thrusts. Afterwards, the harmonic-to-noise ratio (HNR) feature extraction technique was applied to these audio recordings. Finally, model training and validation were performed by applying the Gaussian Naive Bayes machine learning technique to create a diagnostic approach.FindingsA high recall value of 96.19% was obtained in the performance results of the model trained according to damaged and undamaged propeller acoustic data. The precision value was 73.92% as moderate. The overall accuracy value of the model, which can be considered as general performance, was obtained as 81.24%. The F1 score has been found as 83.76% which provides a balanced measure of the model’s precision and recall values.Practical implicationsThis study include provides solid method to diagnose UAV propeller damage using acoustic data obtain from the microphone and allows identification of differently damaged propellers. Using that, the risk of in-flight failures can be reduced and maintenance costs can be lowered with addressing the occurred problems with UAV propeller before they worsen.Originality/valueThis study introduces a novel method to diagnose damaged UAV propellers using the HNR feature extraction technique and Gaussian Naive Bayes classification method. The study is a pioneer in the use of HNR and the Gaussian Naive Bayes and demonstrates its effectiveness in augmenting UAV safety by means of propeller damages. Furthermore, this approach contributes to UAV operational reliability by bridging the acoustic signal processing and machine learning.

Identification of Bolt Loosening Damage of Steel Truss Structure Based on MFCC-WPES and Optimized Random Forest

In the field of bolt loosening detection, although some progress has been made, there are still challenges such as high operational complexity, single feature extraction methods, and insufficient analysis model performance, especially in large steel truss structures, where there is a lack of efficient and accurate bolt loosening identification solutions. In response to these shortcomings, this article proposes an innovative bolt loosening damage recognition method based on sound signals. This method integrates feature extraction techniques of Mel frequency cepstral coefficients (MFCCs) and wavelet packet energy spectra (WPES), and comprehensively characterizes sound signals by constructing MFCC-WPES combined features. Subsequently, the random forest (RF) algorithm optimized by genetic algorithm was used for feature selection and model training, aiming to improve recognition accuracy and robustness. The experimental results show that this method can not only accurately identify bolt loosening signals in steel truss structure bolt loosening detection, but also has strong identification ability for environmental noise. Compared with traditional methods, the proposed solution in this article shows significant improvements in both performance and practicality, providing a new perspective and solution for the technological advancement of bolt loosening detection in steel truss structures.

English Sentiment Analysis and its Application in Translation Based on Decision Tree Algorithm

Sentimental analysis belongs to the class of Natural Language Processing (NLP) based on the rule and machine model. The proposed model comprises of the pre-defined function for the estimation of the features in the English statements. This paper presents the Reflect Sentiment Translation Decision Tree (RSTDT), a novel model designed to integrate sentiment analysis and translation tasks for English text. The RSTDT model combines the strengths of decision tree algorithms with feature extraction techniques to accurately analyze sentiment and translate text across languages. The proposed RSTDT dataset comprises English sentences with annotated sentiment labels, the RSTDT model is trained to identify sentiment polarity and generate corresponding translations in Arabic. The proposed RSTDT model uses Traslation mapping for the estimation of the sentimental features. In order to estimate and classify the features in the neural network, the processes features are assessed using the decision tree model. The RSTDT model's efficacy in precisely capturing sentiment nuances and generating linguistically appropriate translations was shown through thorough testing and review. The model achieves high accuracy in sentiment analysis and exhibits proficiency in translating sentiment-rich content into Arabic while maintaining contextual relevance. Additionally, robust classification performance metrics underscore the model's efficacy in accurately classifying English words into sentiment categories. The RSTDT model offers a promising solution for multilingual sentiment analysis applications, with potential applications in social media monitoring, customer feedback analysis, and cross-cultural sentiment analysis.

Gender classification based on handwriting using Gurmukhi characters and hybrid feature extraction techniques

Gender classification based on handwriting using Gurmukhi characters and hybrid feature extraction techniques

Blood Cancer Identification on Peripheral Blood Smear (PBS) Images Using HSV Feature Extraction

Blood cancer is a category of diseases that have an impact on the development and operation of blood cells. Due to the complexity and diversity of these diseases, proper diagnosis is required before starting treatment. Medical imaging techniques have undergone significant advances in recent years, especially in peripheral blood smear (PBS) image processing. The aim of this study was to uncover how important the extraction of PBS image features is for the diagnosis of blood cancer. Feature extraction is essential to detect anomalies in blood cells in terms of blood cancer detection. The method used is feature extraction based on hue and saturation values (HSV) and uses Machine Support Vector Machine (SVM) machine learning algorithms in classifying malignant and benign PBS images. PBS image data used in this study was 100 images, consisting of 50 malignant PBS images, and 50 benign PBS images. Through the application of HSV feature extraction techniques and PBS image analysis, SVM algorithms can uncover latent indicators of blood cancer and facilitate timely and precise diagnosis. With the SVM technique, classification accuracy can be achieved by 92%. These results demonstrate the potential effectiveness of this feature extraction method. Extraction of HSV features may alter the diagnosis of blood cancer with additional research and application in clinical settings.

Application of Event Cameras and Neuromorphic Computing to VSLAM: A Survey.

Simultaneous Localization and Mapping (SLAM) is a crucial function for most autonomous systems, allowing them to both navigate through and create maps of unfamiliar surroundings. Traditional Visual SLAM, also commonly known as VSLAM, relies on frame-based cameras and structured processing pipelines, which face challenges in dynamic or low-light environments. However, recent advancements in event camera technology and neuromorphic processing offer promising opportunities to overcome these limitations. Event cameras inspired by biological vision systems capture the scenes asynchronously, consuming minimal power but with higher temporal resolution. Neuromorphic processors, which are designed to mimic the parallel processing capabilities of the human brain, offer efficient computation for real-time data processing of event-based data streams. This paper provides a comprehensive overview of recent research efforts in integrating event cameras and neuromorphic processors into VSLAM systems. It discusses the principles behind event cameras and neuromorphic processors, highlighting their advantages over traditional sensing and processing methods. Furthermore, an in-depth survey was conducted on state-of-the-art approaches in event-based SLAM, including feature extraction, motion estimation, and map reconstruction techniques. Additionally, the integration of event cameras with neuromorphic processors, focusing on their synergistic benefits in terms of energy efficiency, robustness, and real-time performance, was explored. The paper also discusses the challenges and open research questions in this emerging field, such as sensor calibration, data fusion, and algorithmic development. Finally, the potential applications and future directions for event-based SLAM systems are outlined, ranging from robotics and autonomous vehicles to augmented reality.

Estimation of Water Velocity and Receiver Position Changes in Time-Lapse Seismic Using Machine Learning

Non-repeatability is one of the main obstacles in time-lapse seismic, as it significantly degrades the interpretation of reservoir-related signals. Correcting the data variations caused by non-repeatability (4D noise) is of paramount importance, which usually requires the estimation of the changing parameter. ln this paper, we propose a Machine Learning (ML) workflow for the quantitative estimation of two types of 4D noise: changes in the speed of sound in water and receiver lateral positions. A synthetic database, modeled from a velocity model estimated from the Brazilian pre-salt, and containing many time-lapse seismic surveys was generated for the supervised training of ML models. lnput samples consist of subsets of common-shot seismograms. We studied many combinations of ML regression algorithms and feature extraction techniques, for scenarios where data is contaminated, or not, with Gaussian random noise. The regression algorithms considered were Fully-Connected Neural Network, Extreme Gradient Boosting, and Bayesian Ridge. Four rectangular crops of the common-shot seismogram were tested as input features: full seismogram, first half of time samples, 11 smallest-offset traces, and the time samples focusing on the first arrivals region. The combination with the best trade-off between accuracy and model complexity is the Bayesian Ridge fed with the 11 smallest-offset traces, which estimated position and velocity time-lapse changes with median accuracy of 0.115 m and 0.017 m/s for the case with Gaussian noise. Besides the correction of repeatability-related variations, our results are useful in the 4D Full-Waveform lnversion which needs accurate parameters to produce good seismic images.

Prediction of short circuit current of wind turbines based on artificial neural network model

The growth of renewable energy on a global scale is making significant strides in power plants. This is due to the increasing concern about climate change, the rising demand for electricity, and the necessity to reduce reliance on fossil fuels. Ensuring the successful integration of new energy resources into the existing network is just as crucial as it requires the system to be reliable and adaptable. For instance, wind energy, which is one of the renewable sources, has an intermittent nature that necessitates the ability to synchronize its actions to achieve the desired system performance. The objective of this study is to utilize a new neural network system to calculate the short circuit current of power plants. Specifically, the focus is on identifying and categorizing the short circuit faults that occur between the stator coils of the squirrel cage induction generator used in wind power generation. To achieve this, a system was developed to simulate turbine data. Subsequently, four feature extraction techniques and machine learning algorithms were employed to enable early detection of short circuit faults. The numerical results obtained from the simulation demonstrated the high efficiency and accuracy of the proposed model.

Monitoring soil nutrients using machine learning based on UAV hyperspectral remote sensing

ABSTRACT Unmanned aerial vehicles (UAV) are rapidly evolving experimental platforms that play an important role in remote sensing. In this study, we investigated a machine learning method for monitoring soil nutrient content using UAV hyperspectral remote sensing. We employed machine learning techniques for feature extraction and soil hyperspectral information modelling. In contrast to traditional mathematical transformation methods, we adopted a combination of random forest and differential evolution algorithms to rank the weights of individual hyperspectral data, thereby obtaining a series of spectral feature subsets for soil organic matter, total nitrogen and available phosphorus and potassium. Furthermore, the analytic hierarchy process was used for weight analysis, and the characteristic bands of the four soil nutrients were successfully extracted. Next, a quantitative inversion model based on a back-propagation (BP) neural network was established to estimate soil nutrient content, with determination coefficients higher than 0.7 and 0.6 for the modelling and verification sets, respectively. The relative percent difference values were greater than 2, among which the highest was for available potassium, with determination coefficients of 0.95 and 0.84 for the modelling and verification sets, respectively. In addition, visualization distribution maps of soil nutrients were obtained by combining the BP model and original reflectance hyperspectral images, and the comparisons of content histograms showed a relatively consistent distribution between the sampling and inversion points. The results verified the effectiveness of the combined machine learning method for large-scale and high-precision monitoring and visualization of soil nutrient contents.

Exploring the impact of nocturnal boundary layer stability on wintertime air pollution in a highly polluted basin city using unsupervised learning classification

This study utilizes ten years of wintertime boundary layer meteorological and surface air quality observations to characterize the nocturnal boundary layer (NBL) stability and assess its relationship with air pollution in Taiyuan, a highly polluted basin city in China. An unsupervised learning feature extraction technique known as the self-organizing map (SOM) is applied to objectively classify nocturnal virtual potential temperature (VPT) profiles. The SOM-based classification scheme allows the representation of wintertime day-to-day NBL evolutions by just nine regimes. Special attention is given to four dominant regimes: weak to moderate stability regime (NBL1), cloudy moderate stability regime (NBL3), windy moderate stability regime (NBL7), and strong stability regime (NBL9). These dominant regimes have relatively higher occurrence frequencies (>10%), with the highest frequency associated with the strong stability regime (NBL9) at 25.2%. The diurnal cycles of selected pollutants (CO, NO2, SO2, and PM2.5) exhibit significant distinctions among the different NBL regimes. For instance, in the strong stability regime, CO, NO2, and PM2.5 show explosive growth in the evening due to the accumulation of primary pollutants. However, in the cloudy moderate stability regime, PM2.5 exhibits persistent slow growth throughout the day, likely due to secondary particle formation under high humidity and high SO2 conditions. These findings enhance our understanding of NBL meteorological impacts on surface air pollution in basin cities.

An optimized method for mulberry silkworm, Bombyx mori (Bombycidae:Lepidoptera) sex classification using TLBPSGA-RFEXGBoost.

Silkworm seed production is vital for silk farming, requiring precise breeding techniques to optimize yields. In silkworm seed production, precise sex classification is crucial for optimizing breeding and boosting silk yields. A non-destructive approach for sex classification addresses these challenges, offering an efficient alternative that enhances both yield and environmental responsibility. Southern India is a hub for mulberry silk and cocoon farming, with the high-yielding double-hybrid varieties FC1 (foundation cross 1) and FC2 (foundation cross 2) being popular. Traditional methods of silkworm pupae sex classification involve manual sorting by experts, necessitating the cutting of cocoons - a practice with a high risk of damaging the cocoon and affecting yield. To address this issue, this study introduces an accelerated histogram of oriented gradients (HOG) feature extraction technique that is enhanced by block-level dimensionality reduction. This non-destructive method allows for efficient and accurate silkworm pupae classification. The modified HOG features are then fused with weight features and processed through a machine learning classification model that incorporates recursive feature elimination (RFE). Performance evaluation shows that an RFE-hybridized XGBoost model attained the highest classification accuracy, achieving 97.2% for FC1 and 97.1% for FC2. The model further optimized with a novel teaching learning-based population selection genetic algorithm (TLBPSGA) achieved a remarkable accuracy of 98.5% for FC1 and 98.2% for FC2. These findings have far-reaching implications for improving both the ecological sustainability and economic efficiency of silkworm seed production.

Deep Learning Approach for Traffic Congestion Sound Classification using Circular Neural Networks

Traffic congestion has become one of the main problems that occur in big cities around the world. Traffic congestion also has a negative impact if not handled seriously. Traffic congestion occurs because there is a buildup of vehicle volume that exceeds the capacity of the road. The efficiency and quality of living in cities can be negatively impacted by traffic congestion, which can also result in higher fuel consumption, pollution, and delays. There needs to be a method that can overcome and identify this. Therefore, by classifying sounds, this research aims to reduce traffic congestion. The author uses deep learning with the Convolutional Neural Network (CNN) method as the algorithm model. The model employs Mel-Frequency Cepstral Coefficients (MFCC) as the primary feature extraction technique to capture the essential characteristics of the audio signals. This research is expected to be able to classify traffic congestion sounds with good accuracy, so it can be used as a solution to overcome traffic congestion. Experiments were conducted using a training dataset, and for testing, the road sound dataset has been collected at traffic light intersections. To evaluate the proposed method, the implementation showed promising results, achieving an accuracy of 97.62% on the training data and 88.19% on the test data in classifying traffic congestion sounds.

Segmentation-Free Outcome Prediction from Head and Neck Cancer PET/CT Images: Deep Learning-Based Feature Extraction from Multi-Angle Maximum Intensity Projections (MA-MIPs).

We introduce an innovative, simple, effective segmentation-free approach for survival analysis of head and neck cancer (HNC) patients from PET/CT images. By harnessing deep learning-based feature extraction techniques and multi-angle maximum intensity projections (MA-MIPs) applied to Fluorodeoxyglucose Positron Emission Tomography (FDG-PET) images, our proposed method eliminates the need for manual segmentations of regions-of-interest (ROIs) such as primary tumors and involved lymph nodes. Instead, a state-of-the-art object detection model is trained utilizing the CT images to perform automatic cropping of the head and neck anatomical area, instead of only the lesions or involved lymph nodes on the PET volumes. A pre-trained deep convolutional neural network backbone is then utilized to extract deep features from MA-MIPs obtained from 72 multi-angel axial rotations of the cropped PET volumes. These deep features extracted from multiple projection views of the PET volumes are then aggregated and fused, and employed to perform recurrence-free survival analysis on a cohort of 489 HNC patients. The proposed approach outperforms the best performing method on the target dataset for the task of recurrence-free survival analysis. By circumventing the manual delineation of the malignancies on the FDG PET-CT images, our approach eliminates the dependency on subjective interpretations and highly enhances the reproducibility of the proposed survival analysis method. The code for this work is publicly released.