AI Assisted Nano Biosensor System for Extra Virgin Olive Oil Adulteration Detection

AI Assisted Nano Biosensor System for Extra Virgin Olive Oil Adulteration Detection Prem P1,*, Nageswari G2, Nikki John Kannampilly3, Hema Prabha P4, Premkumar J4 1Department of Electrical and Electronics Engineering, Ramco Institute of Technology, Rajapalayam, TN, India. 2Department of Food Technology, Sri Shakthi Institute of Engineering and Technology, Coimbatore, TN, India. 3Department of Food Technology, Amal Jyothi College of Engineering, Kottayam, Kerala. 4Department of Food Technology, Nehru Institute of Technology, Coimbatore, TN, India. Abstract The authenticity and purity of extra virgin olive oil (EVOO) are frequently compromised by adulteration with lowergrade or cheaper oils, posing significant risks to consumer health and market integrity. This study introduces an AIassisted nano-biosensor system that combines the high sensitivity of carbon nanotube (CNT)-based sensing elements with the analytical intelligence of a Convolutional Neural Network (CNN) model for rapid, real-time adulteration detection. The CNTs, functionalized with selective bio molecular probes, capture distinct dielectric and electrochemical signatures from EVOO samples. These responses are analyzed using a CNN trained on multi-modal spectral datasets, enabling accurate feature extraction and classification. Experimental evaluations demonstrate that the proposed hybrid nano-sensor–AI system achieves a remarkable classification accuracy of 98.6%, with sensitivity and specificity values of 97.2% and 99.1%, respectively, effectively identifying adulteration levels as low as 1%. Simulated system responses further validate its capability for fast, non-destructive, and intelligent assessment of EVOO quality. The developed system represents a promising advancement in smart food authentication technologies, offering a reliable and scalable approach for ensuring purity, preventing fraud, and strengthening consumer trust in high-value edible oils. Keywords: Nano-biosensor, Carbon nanotube, Extra virgin olive oil, Convolutional neural network, Adulteration detection. I. INTRODUCTION Extra virgin olive oil (EVOO) is one of the most prized edible oils, renowned for its rich nutritional profile, unique sensory attributes, and numerous health benefits arising from its high content of monounsaturated fatty acids and bioactive compounds such as polyphenols and tocopherols. As global demand continues to rise, the authenticity of EVOO has become a significant concern due to its frequent adulteration with lower-cost vegetable oils such as sunflower, canola, soybean, or refined olive oils. Such adulteration not only diminishes the nutritional and sensory quality of the product but also undermines consumer trust, disrupts fair market trade, and may pose potential health risks to consumers. Traditional analytical methods employed for detecting adulteration such as gas chromatography (GC), high-performance liquid chromatography (HPLC), mass spectrometry (MS), and Fourier-transform infrared spectroscopy (FTIR) offers high precision and reliability. However, these techniques are typically time-intensive, require complex sample preparation, involve costly instrumentation, and must be operated by trained professionals in laboratory environments [1]. Consequently, there is a growing demand for rapid, portable, Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-720 and cost-effective analytical tools capable of providing real-time quality assessment of EVOO [2]. Nano-biosensor technology has emerged as a promising alternative, offering enhanced sensitivity, selectivity, and miniaturization potential. Among various nanomaterials, carbon nanotubes (CNTs) have demonstrated exceptional potential for biosensing applications due to their high electrical conductivity, mechanical strength, chemical stability, and large surface area available for functionalization with bio molecular probes [3]. These properties enable CNT-based sensors to capture subtle dielectric, electrochemical, or optical variations induced by adulterants in EVOO samples. In recent years, the integration of artificial intelligence (AI), particularly deep learning techniques such as Convolutional Neural Networks (CNNs), has revolutionized data interpretation in sensor-based systems. CNNs excel at identifying hidden patterns and nonlinear relationships within high-dimensional datasets, making them ideal for analyzing complex spectral or electrochemical responses generated by nano-biosensors [4]. The synergy between nanotechnology and AI has opened a new frontier in intelligent sensing systems capable of autonomous decision-making and real-time classification. This study focuses on the development and simulation of an AI-assisted CNT-based nanobiosensor system for the detection of EVOO adulteration. The proposed hybrid platform leverages CNT-based sensing elements for signal acquisition and a CNN model for automated pattern recognition. By merging advances in material science, sensor electronics, and deep learning, this system aims to achieve a next-generation solution for rapid, reliable, and intelligent authentication of edible oils. The research also explores simulated performance metrics such as classification accuracy, sensitivity and specificity to evaluate the feasibility of deploying such systems for on-site quality assurance and food fraud prevention. II. LITERATURE SURVEY The detection of adulteration in extra virgin olive oil (EVOO) has traditionally relied on chromatographic and spectroscopic techniques such as gas chromatography (GC), highperformance liquid chromatography (HPLC), and mass spectrometry (MS). While these methods offer high precision, they are limited by high operational costs, complex sample preparation, and the requirement of skilled personnel, which restrict their applicability for onsite or real time analysis [5]. Consequently, recent research has shifted toward developing portable, cost effective sensing technologies capable of rapid detection and classification of adulterated oils. Electrochemical biosensors, particularly those incorporating nanomaterials, have gained significant attention due to their miniaturization potential, high sensitivity, and capability for label-free detection [6]. Nanostructured materials such as carbon nanotubes (CNTs), graphene, and gold nanoparticles (AuNPs) have been widely investigated as transducers in electrochemical sensors owing to their excellent electron mobility, high surface area, and ease of surface functionalization. CNT-modified electrodes, in particular, have demonstrated superior electrochemical performance in detecting phenolic compounds and oxidation markers present in EVOO [7]. Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-721 Similarly, grapheneand AuNP-based sensors have been used to enhance charge transfer and improve selectivity in detecting chemical and biological analytes. However, despite these advancements, most nanomaterial-based sensing platforms still depend on conventional chemometric or regression-based models, which are often sensitive to noise and limited in their capacity to model non-linear relationships between sensor signals and oil adulteration levels [8]. To overcome these challenges, recent studies have integrated artificial intelligence (AI), especially deep learning methods such as Deep Neural Networks (DNNs) and Convolutional Neural Networks (CNNs), for analyzing high-dimensional spectral data. These models have shown remarkable improvements in classification accuracy and robustness when applied to Near-Infrared (NIR), Raman, and Fourier Transform Infrared (FTIR) spectroscopy data for food adulteration detection [9]. CNNs, in particular, excel at extracting hierarchical features from complex sensor signals without requiring handcrafted pre-processing steps. The combination of advanced AI models with spectroscopic or electrochemical data has demonstrated superior performance compared to traditional machine learning approaches like Support Vector Machines (SVMs) or Partial Least Squares (PLS) regression, especially under noisy experimental conditions [10]. Despite these promising developments, very few studies have explored the integration of CNT-based electrochemical sensors with AI-driven deep learning models for EVOO adulteration detection. Existing works have primarily focused on optical or spectroscopic data rather than leveraging the electrochemical and dielectric response characteristics of CNTbased nanostructures. The current research addresses this critical gap by proposing a hybrid framework that combines CNT-based nano-biosensing with a CNN classification model. This synergy allows real-time feature extraction and intelligent adulteration detection, achieving high accuracy, sensitivity, and specificity even at low adulteration levels. The proposed system thus represents a significant advancement toward developing portable, intelligent, and reliable solutions for food authenticity verification. III. MATERIALS AND METHODS 3.1 Experimental Design The proposed system combines three core components: 1) CNT based nano-biosensor module 2) Signal acquisition and pre-processing unit 3) AI model (CNN) for classification 3.1.1 Sample Preparation In this research, extra virgin olive oil (EVOO) samples were systematically blended with selected adulterant oils like sunflower, soybean, and palm oils to simulate varying degrees of adulteration typically encountered in commercial practices. Six concentration levels were Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-722 prepared, consisting of 0%, 1%, 5%, 10%, 20%, and 50% (v/v) adulterant compositions. The 0% sample served as the pure EVOO control, while the remaining mixtures represented progressively adulterated samples. All oil samples were mixed thoroughly under controlled laboratory conditions to ensure homogeneous blending and minimize phase separation. Each prepared mixture was subjected to simulated electrochemical impedance spectroscopy (EIS) and dielectric characterization to obtain distinct electrical and physicochemical signatures corresponding to varying adulteration levels. The EIS spectra captured variations in charge transfer resistance, doublelayer capacitance, and interfacial polarization effects, which are indicative of compositional and molecular alterations caused by adulteration. Fig. 1 Simulated Electrochemical Impedance Spectroscopy (EIS) of the Prepared Mixture The fig. 1 illustrates simulated Electrochemical Impedance Spectroscopy (EIS) results for pure and adulterated Extra Virgin Olive Oil (EVOO) samples using a CNT-based nanobiosensor interface. The Nyquist plot (left) shows semi-circular arcs representing impedance behavior, where the diameter corresponds to the charge transfer resistance (Rct). Pure EVOO exhibits the largest arc, indicating higher Rct and stable dielectric properties, while increasing adulteration (10–50%) reduces the arc diameter, signifying enhanced charge transfer and higher ionic mobility due to compositional impurities. The Bode plot (right) displays the relationship between impedance magnitude, phase angle, and frequency, revealing that adulterated samples exhibit greater phase shifts and attenuation at mid to high frequencies, reflecting altered dielectric relaxation and interfacial behavior. Together, these results provide clear electrochemical fingerprints for different adulteration levels, confirming the sensitivity of EIS in detecting physicochemical changes in EVOO mixtures. These datasets formed the input signals for the AI assisted carbon nanotube (CNT) nanobiosensor model, allowing the convolutional neural network (CNN) to learn discriminative patterns from the raw electrochemical response. This systematic sample preparation ensured reproducibility, precise calibration of the sensor response, and reliable model training for subsequent adulteration detection and classification. Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-723 3.1.2 CNT Nano Biosensor Fabrication A biosensor is an analytical device that combines a biological recognition element with a transducer to detect specific analytes with high sensitivity and specificity. The bioreceptor is the core component responsible for recognizing the target substance, and it can be an enzyme, antibody, nucleic acid, cell, or aptamer, depending on the application. Once the bioreceptor interacts with the analyte, this interaction produces a measurable change, which is captured by the transducer. The transducer converts this biological response into a quantifiable electrical, optical, thermal, or piezoelectric signal. To ensure accurate and stable readings, the system often includes a signal processor that amplifies and filters the signal, followed by a display or output interface that presents the data in a usable form. Fig. 2 shows the schematic of the CNT-based nano-biosensor. Fig. 2 Schematic representation of the carbon nanotube-based nano-biosensor design [11] The CNTs were functionalized with carboxyl (-COOH) and hydroxyl (-OH) groups to enhance analyte interaction. The sensor electrodes were fabricated on a gold interdigitated substrate (IDEs) via chemical vapor deposition. 3.1.3 Signal Acquisition Sensor output signals were captured in the 1 Hz–1 MHz frequency range, digitized at 10-bit resolution, and filtered using a Butterworth low-pass filter. Key parameters extracted included impedance magnitude (|Z|), phase angle (θ), and current response (I). 3.1.4 Dataset Description A total of 1800 simulated sensor readings were generated. Each sample contained 250 features representing spectral intensities. Table I summarizes the dataset structure. Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-724 Table I Summary of the dataset used for CNN training and validation. Dataset Type Samples Features per Sample Adulterant Range (%) Training 1200 250 0–50 Validation 300 250 0–50 Testing 300 250 0–50 3.1.5 CNN Model Architecture A Convolutional Neural Network (CNN) is a type of deep learning model designed to automatically extract and learn hierarchical features from input data, typically images. It begins with an input layer that receives the raw image, followed by one or more convolutional layers, which apply learnable filters to detect local patterns such as edges, textures, and shapes. After each convolution, an activation function like ReLU introduces non-linearity, enabling the network to model complex relationships. Pooling layers then reduce the spatial dimensions of the feature maps, lowering computational load and improving translation invariance. Stacking multiple convolutional and pooling layers allows the network to learn increasingly abstract features. The flatten layer converts the final feature maps into a one-dimensional vector, which is fed into fully connected (dense) layers for highlevel reasoning and classification. Finally, the output layer, often uses Softmax activation, produces the probability distribution over the target classes, enabling accurate predictions based on the learned features. This architecture in fig. 3 makes CNNs highly effective for image recognition, object detection, and other computer vision tasks. Fig. 3 Schematic diagram of basic convolutional neural network (CNN) architecture [12] Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-725 A CNN model was designed with two convolutional layers (kernel size = 3×3), one pooling layer, and two fully connected layers. Fig. 4 illustrates the CNN architecture for EVOO adulteration classification. Fig. 4 CNN architecture for EVOO adulteration classification 3.1.6 Model Training The model was trained using the Adam optimizer with a learning rate of 0.001, batch size of 32, and 100 epochs. Early stopping was applied to prevent overfitting. The loss and accuracy curves during training are shown in Fig. 5. Fig. 5 Training and validation accuracy and loss curves for the CNN model Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-726 IV. RESULTS AND DISCUSSION 4.1 Model Performance The confusion matrix illustrates the classification performance of the proposed Convolutional Neural Network (CNN) model for detecting adulteration in extra virgin olive oil (EVOO) samples. The matrix displays the number of correctly and incorrectly classified samples across two categories like Pure and Adulterated. The diagonal elements (50 and 46) represent the true positive and true negative predictions, indicating correctly classified samples. The off-diagonal elements (5 and 4) correspond to misclassifications, where pure samples were incorrectly identified as adulterated and vice versa. Fig. 6 Confusion matrix for CNN classification results From the fig. 6, it is evident that the CNN achieved a high overall accuracy, with a strong concentration of values along the diagonal. This demonstrates the model’s effective learning capability in distinguishing between authentic and adulterated EVOO samples based on extracted spectral or image-based features. The colour gradient, ranging from light to dark blue, visually indicates the density of correctly classified instances, with darker regions representing higher accuracy. Performance Evaluation of CNN Model Based on the confusion matrix (Fig. 6), the performance of the Convolutional Neural Network (CNN) model for classifying pure and adulterated extra virgin olive oil (EVOO) samples was quantitatively assessed using standard evaluation metrics, including accuracy, precision, recall, and F1-score. The results are summarized in table II. Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-727 Table II Performance Metrics Metric Formula Value (%) Accuracy (TP + TN) / (TP + TN + FP + FN) 96.0 Precision TP / (TP + FP) 92.0 Recall (Sensitivity) TP / (TP + FN) 93.5 F1-Score 2 × (Precision × Recall) / (Precision + Recall) 92.7 Where: • TP (True Positives) = 50 • TN (True Negatives) = 46 • FP (False Positives) = 5 • FN (False Negatives) = 4 Fig. 7 Performance metrics of CNN model The CNN model achieved an overall classification accuracy of 96%, indicating robust generalization and minimal overfitting (Fig. 7). The high precision and recall values confirm that the model effectively differentiates between pure and adulterated oil samples with minimal false predictions. The balanced F1-score (92.7%) further reflects the model’s strong reliability for real-world adulteration detection applications. Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-728