AKAnusuya Krishnan

Anusuya Krishnan

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Anusuya Krishnan has a total of 25 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Robust fault classification in rotary machines using recurrence quantification analysis features for machine learning techniques

    by Ayham Zaitouny, Anusuya Krishnan, Houssam Abdul-Rahman, Yaman Sabsabi, Yousef Salah, Osama Al Mukbel, Rafat Damseh, Nazar Zaki

    Fault detection in rotatory machines is an important necessity to provide system operational reliability and lower unexpected downtime in industrial systems. Conventional feature extraction methods, although good in controlled situations, tend to fall short in noisy situations that mirror real-world operation situations. To fulfill this gap, this work explores machine learning classifiers—Random Forest (RF), Gradient Boosting (GB), XGBoost (XGB), and LightGBM (LGBM)—with two different feature sets: statistical features and Recurrence Quantification Analysis (RQA) features. First, we systematically validate the proposed framework by benchmarking recurrence-based dynamical features against conventional statistical features on a synthetic nonlinear benchmark composed of four well-established dynamical systems: the Lorenz system, the Rössler system, the Hénon map, and the Duffing oscillator. This controlled evaluation enables a methodological comparison across distinct nonlinear regimes before applying the framework to industrial fault data. Then, using experimental rotary machines data, we further examine the robustness of the classifiers under two different operation situations: noise-free and noise-included situations. In a noise-free situation, all classifiers resulted near perfect classification with all statistical features obtaining 99%, whereas RQA features resulted similarly around 93%. Yet, when a Gaussian noise is added (15%–75% of signal magnitude), statistical features marked a considerable decline to 85% accuracy, whereas RQA features remained highly efficient above 94%; consistent results were obtained when other types of noise were added, namely, Brownian noise and impulsive noise. These results present a demonstration of RQA-based feature superiority in long-term dynamics capture and cite its potential to provide a more reliable fault detection in industrial practice where the use of noise and variability of environments is an inescapable consequence.

  • Open Access

    Glycemic Pattern Identification in Young Children With Presymptomatic Type 1 Diabetes: Machine Learning Analysis of Continuous Glucose Monitoring Data (Preprint)

    by Ayham Zaitouny, Anusuya Krishnan, Jonathan Chee, Alexandra Tully, Megan As Penno, John M Wentworth, Tony Huynh, Maria E Craig, Peter G Colman, Georgia Soldatos, et al.

    BACKGROUND Continuous glucose monitoring (CGM) of children with early-stage type 1 diabetes (T1D) identifies early glycemic anomalies associated with glycemic progression. This study applied machine learning on CGM features to identify glycemic patterns that precede stage 3 T1D. OBJECTIVE The aim of this study was to apply symbolic transformation and machine learning–based multilevel clustering approaches to cross-sectional CGM data from children with presymptomatic type 1 diabetes to identify glycemic patterns associated with progression to stage 3 T1D. METHODS Cross-sectional CGM data from 79 participants being longitudinally followed in the Environmental Determinants of Islet Autoimmunity (ENDIA) study were analyzed using multilevel clustering frameworks applied at both daily and participant levels. Input features included time spent in glycemic ranges, area under the curve (AUC), average glycemic distribution, and glycemic variability metrics. RESULTS Clustering of daily CGM profiles using k-means identified subgroups characterized by different proportions of CGM time in tight range, hyperglycemia and/or hypoglycemia. In parallel, participant level clustering was performed to identify subgroups of individual glycemic profiles consisting of multiple consecutive days. Six of the seven participants who progressed to stage 3 T1D within the study period were identified by one or more clustering approaches. CONCLUSIONS Mathematical models and machine learning approaches to time-series data may provide novel insights from CGM data. A combination of symbolic transformation with statistical clustering at multiple temporal resolutions provides a robust framework for characterizing glycemic variability in CGM data. This exploratory approach could be applied to larger datasets to support early risk stratification and inform the development of CGM-based clinical surveillance strategies.

  • Open Access

    Leveraging Machine Learning Regression Algorithms to Predict Mechanical Properties of Evaporitic Rocks From Their Physical Attributes

    by Ayham Zaitouny, Hasan Arman, Anusuya Krishnan, Alaa Ahmed, Ahmed Gad

    Evaluating the geotechnical properties of evaporitic rocks is crucial for infrastructure stability; however, traditional methods are costly and labour-intensive. In this study, machine learning (ML) regression algorithms were applied to predict four key mechanical parameters, namely, uniaxial compressive strength (UCS), point load index (PLI), indirect tensile strength (ITS), and Schmidt hardness value (SHV), based on the physical attributes of evaporitic rocks. A comprehensive laboratory analysis of 149 block samples from Abu Dhabi was performed to measure their physical (density, porosity, unit weight, water content, specific gravity, and void ratio) and mechanical properties. Nine ML models were trained (80:20 data split) and validated using R-squared (R²), mean absolute error (MAE), and root-mean square error (RMSE). Nonlinear correlations demonstrated strong relationships between the mechanical properties and physical attributes, such as saturated density (s) and natural unit weight (n). After feature selection, random forest and XGBoost outperformed the other models with exceptional accuracy for UCS prediction (R2 = 0.95) and robust performance for ITS (R2 = 0.84) and SHV (R2 = 0.77). This study advances previous research by simultaneously predicting multiple mechanical properties, offering an efficient alternative to expensive laboratory tests. The study findings highlight the potential of ML for improving geotechnical workflows by enabling rapid data-driven examinations of rock durability and stability.


  • Open Access

    Effects of Different Noise Reduction Deep Learning Strategies on Computed Tomography Images

    by Zsolt Adam Balogh, Anusuya Krishnan, Mahmoud Nizar Hassan

    Noise reduction in Computed Tomography (CT) is very important for two reasons. On one hand, it can improve the quality of reconstructions, and on the other hand, significant dose reduction can be achieved, which is important due to the radiation risk to the patient. In this area, deep learning-based noise reduction possibilities are in the focus of interest. There are numerous different network models with various training strategies. In this paper, we examined different supervised training strategies with four different deep-learning architectures, and for the training, we only used mathematical phantoms with artificially simulated noise. The noise model was designed for a real multi-slice cone-beam CT and was performed in projection space on forward-projected mathematical phantoms. We validated the noise model and evaluated the effectiveness of different deep learning-based noise reduction models. We redesigned some of the deep learning models and analyzed their noise reduction effect using the noise power spectrum. We examined the noise reduction performance of the models on both real multi-slice cone beam CT scans and mathematical phantoms, with particular attention to their practical applicability.


  • Open Access

    Analyzing the Impact of Augmentation Techniques on Deep Learning Models for Deceptive Review Detection: A Comparative Study

    by Anusuya Kri̇shnan, Kennedyraj Mari̇afranci̇s

    Deep Learning has brought forth captivating applications, and among them, Natural Language Processing (NLP) stands out. This study delves into the role of the data augmentation training strategy in advancing NLP. Data augmentation involves the creation of synthetic training data through transformations, and it is a well-explored research area across various machine learning domains. Apart from enhancing a model's generalization capabilities, data augmentation addresses a wide range of challenges, such as limited training data, regularization of the learning objective, and privacy protection by limiting data usage. The objective of this study is to investigate how data augmentation improves model accuracy and precise predictions, specifically using deep learning-based models. Furthermore, the study conducts a comparative analysis between deep learning models without data augmentation and those with data augmentation.