Optimized Support Vector Machine Research Articles

Optimized Support Vector Machine Using Whale Optimization Algorithm for Acute Lymphoblastic Leukemia Detection from Microscopic Blood Smear Images

Optimized Support Vector Machine Using Whale Optimization Algorithm for Acute Lymphoblastic Leukemia Detection from Microscopic Blood Smear Images

A form-grinding wheel optimization method for helical gears based on a PSO-SVM model

A gear form-grinding optimization method is proposed to obtain an optimal profile of a form-grinding wheel by solving a transcendental equation of the contact line between a form-grinding wheel and a helical gear workpiece. Since the equation of the contact line is transcendental, the relationship between the installation angle of the form-grinding wheel and the shape of the contact line cannot be represented by explicit functions, which makes it difficult to obtain the optimal profile of the form-grinding wheel. An optimization method of the contact line between the form-grinding wheel and the gear is proposed using three evaluation parameters that are the overrun, the shift, and the offset. Some gear form-grinding performances, such as machining stroke, tooth deviation, and grinding chatter, can be quantitatively described using those evaluation parameters. Further, a method for optimizing evaluation parameters is proposed, which uses a particle swarm optimization-support vector machine (PSO-SVM) model with the advantage of small-sample robustness to solve evaluation parameters. The PSO-SVM model is trained under the condition of different installation angles of the form-grinding wheel as the input and the evaluation function as the output. The R-squared value of the overrun is 0.986 that vilidate the high accuracy of the PSO-SVM model. Gear form-grinding test results show the proposed method can effectively improve grinding accuracy.

Short-term wind power output prediction using hybrid-enhanced seagull optimization algorithm and support vector machine: A high-precision method

ABSTRACT Affected by the uncertainty of external environmental factors, wind power generation has significant characteristics of randomness and non-stationarity. Accurately predicting wind power is a necessary technical means to improve the efficiency of wind energy utilization and reduce wind abandonment. Integrating large-scale wind power into the grid has put higher requirements for wind power prediction accuracy. Hence, this study proposes a combined prediction model based on wavelet denoising (WD), hybrid enhanced seagull optimization algorithm (HESOA) and support vector machine (SVM) to predict short-term wind power. The WD technology obtains effective signals from original wind power data, providing a reasonable basis for prediction research. The HESOA is proposed by introducing the Tent mapping, individual self-learning factor and Gaussian population strategy into the original seagull optimization algorithm to optimize SVM’s prediction ability. The case study results reflect that the prediction method is feasible and universal in suppressing wind power generation volatility in spring and winter. The RMSE of short-term wind power prediction is controlled below 65 kW, and the value of R2 is above 98%. This study improves the accuracy of wind power prediction and helps to strengthen the renewable energy consumption level and reliability of wind power grid-connected generation systems.

Multi-objective optimization of the performance for a marine methanol-diesel dual-fuel engine

Marine methanol-diesel compound combustion (DMCC) engines have received extensive attention in the shipping industry. This work aims to further improve the performance and evaluate the carbon reduction potential of the marine DMCC engine. First, the computational fluid dynamic model of the dual-fuel engine was built. Then, an intelligent regression method based on the grey wolf optimizer (GWO) algorithm optimized support vector machine regression (SVR) was proposed. The methanol substitution rate (MSR), excess air ratio (EAR), and diesel injection timing (DIT) were adopted as the decision variables to establish the regression models for NOx, CO, and CO2 emissions and indicated specific fuel consumption (ISFC). Subsequently, a multi-objective grey wolf optimizer (MOGWO) and entropy weight TOPSIS were employed to obtain the optimal case for achieving low NOx and CO emissions and ISFC. Finally, the maximum MSR was explored to achieve greater carbon reduction benefits while balancing the NOx emission and ISFC. The results showed that after GWO optimization, the coefficients of determination of the SVR regression models are all greater than 0.95, with mean square errors consistently below 0.015. It indicates that the regression models have excellent consistency and applicability. The optimal case calculated by MOGWO and entropy-weighted TOPSIS enables co-optimization of the NOx emission and ISFC while maintaining the CO emission below 600 ppm. Compared to the original case, the optimal case achieves reductions of 26.22 % for NOx and 13.39 % for ISFC. The exploration of the maximum MSR reveals that in the Pareto optimal set, the maximum MSR reaches 56.16 %, at which point the NOx and CO2 emissions reduce by 75.85 % and 59.32 % respectively, while the ISFC increases by 21.67 % compared to the original case.

Predicting Onset of Sand Production in Oil Wells using Machine Learning

Sand production in oil wells is a significant challenge that negatively impacts productivity and compromise equipment integrity. This study explores the application of Optimized Support Vector Machine (SVM) binary classification algorithm to predict the onset of sand production in oil wells. A dataset from 63 oil wells was utilized, and class labels were determined based on the bulk and shear modulus product. The model development incorporated geological and mechanical parameters that could influence sand detachment such as: Young’s modulus, Poisson’s ratio, minimum and maximum horizontal stresses, overburden pressure, pore pressure, depth, fracture gradient, and formation strength. Instances above the threshold of 8E+11 were classified as indicative of no sand production, while those below were considered potential sand production scenarios. The SVM model demonstrated remarkable accuracy in predicting sand production onset, trained and tested rigorously with field data. The model's accuracy was evaluated using statistical parameters, such as: accuracy (ACC), sensitivity (SE), specificity (SP), and Matthew's Correlation Coefficient (MCC). From the results, the model achieved a score of 1 across all parameters, indicating high reliability and accuracy in sand production prediction. The practical implications of this model are significant, offering assistance to completion engineers in making proactive decisions regarding sand control strategies. Furthermore, the integration of this model into oil and gas industry processes can optimize operational efficiency by foreseeing potential sand production events, hence, preventing production impairment and ensuring loss prevention.

A rolling bearing fault diagnosis technique based on recurrence quantification analysis and Bayesian optimization SVM

A rolling bearing fault diagnosis technique is proposed based on Recurrence Quantification Analysis (abbreviated as RQA) and Bayesian optimized Support Vector Machine (abbreviated as RQA-Bayes-SVM). Firstly, analyzing the vibration signal with recurrence plot and the nonlinear feature parameters are extracted with RQA, constructing a feature matrix describing the fault mode and fault degree comprehensively. Finally, Bayesian optimization algorithm is introduced for searching the best penalty factor C and kernel function parameter g of SVM and establishing an optimal Bayes-SVM model. Bearing datasets from CWRU is imported for diagnosis on fault mode and fault degree. The results show that the technique presents a good performance on fault mode diagnosis as well as fault degree distinction. Compared with common k-Nearest Neighbor (abbreviated as KNN) and Random Forest (abbreviated as RF) diagnosis models, Bayes-SVM has the best accuracy and stability, which indicates a potential value for engineering applications.

Acoustic tunnel lining cavity detection using cepstral coefficients with optimized filter bank

Tunnels are an essential component of modern transportation infrastructure, and their structural health is critical to traffic safety, which can be seriously affected by tunnel lining cavities. In this paper, an acoustic-based detection approach for assessing the integrity of tunnel linings is studied. By tapping the tunnel lining surface, acoustic signals are sampled and analyzed using a novel feature parameter extraction algorithm-the energy-frequency cepstral coefficient, which uses wavelet packet decomposition to obtain energy distribution statistics in the frequency domain of the signal, and constructs a signal-dependent filter bank to achieve the cepstral coefficient extraction. Compared with the traditional Mel filter bank, this method can adaptively adjust the resolution of the filter bank according to the frequency characteristics of the classified samples. This allows for higher frequency resolution in regions where the energy distribution is concentrated. As a result, the extracted feature parameters achieve both dimensional compression and superior information retention. Experimental results show that the proposed energy-frequency cepstral coefficient feature outperforms the traditional Mel-frequency cepstral coefficient feature, resulting in a higher accuracy of tunnel lining detection. The convolutional neural network model achieves an accuracy of 99.2%, with a 78.9% reduction in error rate compared with the traditional Mel-frequency cepstral coefficient feature parameters. Additionally, a particle swarm optimization support vector machine model is trained to achieve an accuracy rate of 99.6% and an error rate reduction of 76.5%.

A Bilateral Assessment of Human Activities Using PSO-Based Feature Optimization and Non-linear Multi-task Least Squares Twin Support Vector Machine

A Bilateral Assessment of Human Activities Using PSO-Based Feature Optimization and Non-linear Multi-task Least Squares Twin Support Vector Machine

Methods for Identifying Effective Microseismic Signals in a Strong-Noise Environment Based on the Variational Mode Decomposition and Modified Support Vector Machine Models

The environment for acquiring microseismic signals is always filled with complex noise, leading to the presence of abundant invalid signals in the collected data and greatly disturbing effective microseismic signals. Regarding the identification of effective microseismic signals with a low signal-to-noise ratio, a method for identifying effective microseismic signals in a strong-noise environment by using the variational mode decomposition (VMD) and genetic algorithm (GA)-based optimized support vector machine (SVM) model is proposed. Microseismic signals with a low signal-to-noise ratio are adaptively decomposed into several intrinsic mode functions (IMFs) by using VMD. The characteristics of such IMFs are extracted and used as a basis for the determination of signal validity. The SVM model is optimized by utilizing GA to obtain the optimal penalty factor c and the kernel function parameter g. The availability of IMF components is judged by the optimized SVM model, based on which the effectiveness of microseismic signals is further identified. By applying the algorithm to the microseismic signals with artificially added noise, the effective microseismic signals and ineffective noise are discriminated, verifying the feasibility of the algorithm. After processing the microseismic records collected in the field, we effectively judge the effectiveness of microseismic signals, suppress the interfering noise in the data and greatly improve the signal-to-noise ratio of the seismic records. The results show that the method for identifying effective microseismic signals based on VMD and GA-SVM can well discriminate between effective and ineffective microseismic signals, which is very significant and provides technical support for microseismic monitoring in a strong-noise environment.

Robust Breast Cancer Prognosis Prediction: Adaptive Outlier Removal using SVM and K-Means Clustering

Analyzing several datasets is essential to breast cancer research in order to find trends and prognostic markers. For this reason, the Wisconsin Prognostic Breast Cancer (WPBC) dataset offers a valuable source of data. Outliers, however, have the potential to seriously affect how accurate predictive models are. This work suggests using the Support Vector Machine (SVM) algorithm in an adaptive outlier removal method to improve the resilience of prediction models that were trained on the WPBC dataset. To ensure optimum SVM performance, the technique includes pre-processing processes, including addressing missing data and standardizing features. Tailored elimination of outliers is made possible by their dynamic identification, depending on how they deviate from the support of the SVM model. To increase generalization, the SVM is then retrained using the outlier-adjusted dataset. Test set evaluation shows the effectiveness of the method with improved F1-score, recall, and accuracy. With datasets similar to WPBC, this adaptive outlier elimination technique offers a useful tool for improving breast cancer prediction models, leading to increased model performance and dependability in prognostic tasks.

Data-augmented landslide displacement prediction using generative adversarial network

Landslides are destructive natural disasters that cause catastrophic damage and loss of life worldwide. Accurately predicting landslide displacement enables effective early warning and risk management. However, the limited availability of on-site measurement data has been a substantial obstacle in developing data-driven models, such as state-of-the-art machine learning (ML) models. To address these challenges, this study proposes a data augmentation framework that uses generative adversarial networks (GANs), a recent advance in generative artificial intelligence (AI), to improve the accuracy of landslide displacement prediction. The framework provides effective data augmentation to enhance limited datasets. A recurrent GAN model, RGAN-LS, is proposed, specifically designed to generate realistic synthetic multivariate time series that mimics the characteristics of real landslide on-site measurement data. A customized moment-matching loss is incorporated in addition to the adversarial loss in GAN during the training of RGAN-LS to capture the temporal dynamics and correlations in real time series data. Then, the synthetic data generated by RGAN-LS is used to enhance the training of long short-term memory (LSTM) networks and particle swarm optimization-support vector machine (PSO-SVM) models for landslide displacement prediction tasks. Results on two landslides in the Three Gorges Reservoir (TGR) region show a significant improvement in LSTM model prediction performance when trained on augmented data. For instance, in the case of the Baishuihe landslide, the average root mean square error (RMSE) increases by 16.11%, and the mean absolute error (MAE) by 17.59%. More importantly, the model's responsiveness during mutational stages is enhanced for early warning purposes. However, the results have shown that the static PSO-SVM model only sees marginal gains compared to recurrent models such as LSTM. Further analysis indicates that an optimal synthetic-to-real data ratio (50% on the illustration cases) maximizes the improvements. This also demonstrates the robustness and effectiveness of supplementing training data for dynamic models to obtain better results. By using the powerful generative AI approach, RGAN-LS can generate high-fidelity synthetic landslide data. This is critical for improving the performance of advanced ML models in predicting landslide displacement, particularly when there are limited training data. Additionally, this approach has the potential to expand the use of generative AI in geohazard risk management and other research areas.

Optimizing Support Vector Machine Performance for Parkinson's Disease Diagnosis Using GridSearchCV and PCA-Based Feature Extraction

Background: Parkinson's disease (PD) is a critical neurodegenerative disorder affecting the central nervous system and often causing impaired movement and cognitive function in patients. In addition, its diagnosis in the early stages requires a complex and time-consuming process because all existing tests such as electroencephalography or blood examinations lack effectiveness and accuracy. Several studies explored PD prediction using sound, with a specific focus on the development of classification models to enhance accuracy. The majority of these neglected crucial aspects including feature extraction and proper parameter tuning, leading to low accuracy. Objective: This study aims to optimize performance of voice-based PD prediction through feature extraction, with the goal of reducing data dimensions and improving model computational efficiency. Additionally, appropriate parameters will be selected for enhancement of the ability of the model to identify both PD cases and healthy individuals. Methods: The proposed new model applied an OpenML dataset comprising voice recordings from 31 individuals, namely 23 PD patients and 8 healthy participants. The experimental process included the initial use of the SVM algorithm, followed by implementing PCA for feature extraction to enhance machine learning accuracy. Subsequently, data balancing with SMOTE was conducted, and GridSearchCV was used to identify the best parameter combination based on the predicted model characteristics. Result: Evaluation of the proposed model showed an impressive accuracy of 97.44%, sensitivity of 100%, and specificity of 85.71%. This excellent result was achieved with a limited dataset and a 10-fold cross-validation tuning, rendering the model sensitive to the training data. Conclusion: This study successfully enhanced the prediction model accuracy through the SVM+PCA+GridSearchCV+CV method. However, future investigations should consider an appropriate number of folds for a small dataset, explore alternative cross-validation methods, and expand the dataset to enhance model generalizability. Keywords: GridSearchCV, Parkinson Disaese, SVM, PCA, SMOTE, Voice/Speech

Fault diagnosis method using MVMD signal reconstruction and MMDE-GNDO feature extraction and MPA-SVM

To achieve a comprehensive and accurate diagnosis of faults in rolling bearings, a method for diagnosing rolling bearing faults has been proposed. This method is based on Multivariate Variational Mode Decomposition (MVMD) signal reconstruction, Multivariate Multiscale Dispersion Entropy (MMDE)-Generalized Normal Distribution Optimization (GNDO), and Marine predators’ algorithm-based optimization support vector machine (MPA-SVM). Firstly, by using a joint evaluation function (energy*|correlation coefficient|), the multi-channel vibration signals of rolling bearings after MVMD decomposition are denoised and reconstructed. Afterward, MMDE is applied to fuse the information from the reconstructed signal and construct a high-dimensional fault feature set. Following that, GNDO is used to select features and extract a subset of low-dimensional features that are sensitive and easy to classify. Finally, MPA is used to realize the adaptive selection of important parameters in the SVM classifier. Fault diagnosis experiments are carried out using datasets provided by the Case Western Reserve University (CWRU) and Paderborn University (PU). The MVMD signal reconstruction method can effectively filter out the noise components of each channel. MMDE-GNDO can availably mine multi-channel fault features and eliminate redundant (or interference) items. The MPA-SVM classifier can identify faults in different working conditions with an average accuracy of 99.72% and 100%, respectively. The results demonstrate the accuracy, efficiency, and stability of the proposed method.

Developing a novel image marker to predict the clinical outcome of neoadjuvant chemotherapy (NACT) for ovarian cancer patients

ObjectiveNeoadjuvant chemotherapy (NACT) is one kind of treatment for advanced stage ovarian cancer patients. However, due to the nature of tumor heterogeneity, the clinical outcomes to NACT vary significantly among different subgroups. Partial responses to NACT may lead to suboptimal debulking surgery, which will result in adverse prognosis. To address this clinical challenge, the purpose of this study is to develop a novel image marker to achieve high accuracy prognosis prediction of NACT at an early stage. MethodsFor this purpose, we first computed a total of 1373 radiomics features to quantify the tumor characteristics, which can be grouped into three categories: geometric, intensity, and texture features. Second, all these features were optimized by principal component analysis algorithm to generate a compact and informative feature cluster. This cluster was used as input for developing and optimizing support vector machine (SVM) based classifiers, which indicated the likelihood of receiving suboptimal cytoreduction after the NACT treatment. Two different kernels for SVM algorithm were explored and compared. A total of 42 ovarian cancer cases were retrospectively collected to validate the scheme. A nested leave-one-out cross-validation framework was adopted for model performance assessment. ResultsThe results demonstrated that the model with a Gaussian radial basis function kernel SVM yielded an AUC (area under the ROC [receiver characteristic operation] curve) of 0.806 ± 0.078. Meanwhile, this model achieved overall accuracy (ACC) of 83.3%, positive predictive value (PPV) of 81.8%, and negative predictive value (NPV) of 83.9%. ConclusionThis study provides meaningful information for the development of radiomics based image markers in NACT treatment outcome prediction.

Optimized Support Vector Machine with Particle Swarm Optimization to Improve the Accuracy Amazon Sentiment Analysis Classification

Text mining is a valuable technique that empowers users to gain a deeper understanding of existing textual data, ultimately allowing them to make more informed decisions. One important application of text mining is in the field of sentiment analysis, which has gained significant traction among companies aiming to understand how customers perceive their products and services. In response to this growing need, various research efforts have been made to improve the accuracy of sentiment analysis classification models. The purpose of this article is to discuss a specific approach using the Support Vector Machine (SVM) algorithm, which is often used in machine learning for text classification tasks and then combined with the application of Particle Swarm Optimization (PSO), which optimizes the SVM model parameters to achieve the best classification results. This dynamic combination not only improves accuracy but also enhances the model's ability to efficiently handle large amounts of text data to achieve better results. The research findings highlight the effectiveness of this approach. The application of the SVM algorithm with PSO resulted in an outstanding accuracy performance of 94.92%. The substantial increase in accuracy compared to previous studies shows the promising potential of this methodology. This proves that the SVM algorithm model approach with Particle Swarm Optimization provides good performance.

Rolling bearing fault diagnosis based on multi-domain features and whale optimized support vector machine

Rolling bearing is an important rotating support component in mechanical equipment. It is very prone to wear, defects, and other faults, which directly affect the reliable operation of mechanical equipment. Its running condition monitoring and fault diagnosis have always been a matter of concern to engineers and researchers. A rolling bearing fault diagnosis technique based on multi-domain feature and whale optimization algorithm-support vector machine (MDF-WOA-SVM) is proposed. Firstly, recursive analysis is performed on vibration signal and the recursive features are employed as nonlinear recursive feature vector including recursive rate (RR), deterministic rate (DET), recursive entropy (RE), and diagonal average length (DAL). Then, a comprehensive multi-domain feature vector is constructed by combining three time-domain features including root mean square, variance, and peak to peak. Finally, whale optimization algorithm (WOA) is introduced to optimize the penalty factor C and kernel function parameter g to construct the optimal WOA-SVM model. The rolling bearing datasets of Jiangnan University is employed for instance analysis, and the results show that the 10-CV accuracy of the technique proposed is good with an accuracy of 99%. Compared with recursive features or time-domain features, multi-domain features are more accurate and comprehensive in describing characters of the signal. Some popular supervised learning models are also introduced for comparison including K-nearest neighbor (KNN) and decision tree (DT), and the result shows that the proposed method has a higher accuracy and certain advantages.

Establishment and comparison of in situ detection models for foodborne pathogen contamination on mutton based on SWIR-HSI.

Rapid and accurate detection of food-borne pathogens on mutton is of great significance to ensure the safety of mutton and its products and the health of consumers. The feasibility of short-wave infrared hyperspectral imaging (SWIR-HSI) in detecting the contamination status and species of Escherichia coli (EC), Staphylococcus aureus (SA) and Salmonella typhimurium (ST) contaminated on mutton was explored. The hyperspectral images of uncontaminated and contaminated mutton samples with different concentrations (108, 107, 106, 105, 104, 103 and 102 CFU/mL) of EC, SA and ST were acquired. The one dimensional convolutional neural network (1D-CNN) model was constructed and the influence of structure hyperparameters on the model was explored. The effects of different spectral preprocessing methods on partial least squares-discriminant analysis (PLS-DA), support vector machine (SVM) and 1D-CNN models were discussed. In addition, the feasibility of using the characteristic wavelength to establish simplified models was explored. The best full band model was the 1D-CNN model with the convolution kernels number of (64, 16) and the activation function of tanh established by the original spectra, and its accuracy of training set, test set and external validation set were 100.00, 92.86 and 97.62%, respectively. The optimal simplified model was genetic algorithm optimization support vector machine (GA-SVM). For discriminating the pathogen species, the accuracies of SVM models established by full band spectra preprocessed by 2D and all 1D-CNN models with the convolution kernel number of (32, 16) and the activation function of tanh were 100.00%. In addition, the accuracies of all simplified models were 100.00% except for the 1D-CNN models. Considering the complexity of features and model calculation, the 1D-CNN models established by original spectra were the optimal models for pathogenic bacteria contamination status and species. The simplified models provide basis for developing multispectral detection instruments. The results proved that SWIR-HSI combined with machine learning and deep learning could accurately detect the foodborne pathogen contamination on mutton, and the performance of deep learning models were better than that of machine learning. This study can promote the application of HSI technology in the detection of foodborne pathogens on meat.

Multivariate multi-scale cross-fuzzy entropy and SSA-SVM-based fault diagnosis method of gearbox

Abstract Fuzzy entropy (FuzzyEn) is widely recognized as a powerful tool for analyzing nonlinear dynamics and measuring the complexity of time series data. It has been utilized as an effective indicator to capture nonlinear fault features in gearbox vibration signals. However, FuzzyEn only measures complexity at a single scale, ignoring the valuable information contained in large-scale features of the time series. Furthermore, FuzzyEn does not account for coupling characteristics between related or synchronized time series. To address these limitations, a novel entropy-based approach called multivariate multi-scale cross-fuzzy entropy (MvMCFE) is proposed in this paper for measuring the complexity and mutual predictability of two multivariate time series. Relying on the advantages of MvMCFE in nonlinear feature extraction, a new fault diagnosis method for gearboxes is proposed based on MvMCFE and an optimized support vector machine (SVM) using the salp swarm algorithm (SSA-SVM). Ultimately, the proposed gearbox diagnostic method is employed to analyze the gearbox experimental data and a comparison with existing fault diagnosis approaches is conducted. The comparison results indicate that the proposed method can effectively extract nonlinear fault features of gearboxes and achieve the highest recognition rate compared to the other methods.

Support vector machine based fault detection in inverter‐fed electric vehicle

AbstractInverters play a prominent role in the power train system of electric vehicles (EVs). Devices in EV connected power system are threatened by faults due to the continuous working and varying speed range of motors in EVs. Hence, in the EV connected application, the detection of fault is essential since it secures the system from severe damage and dangerous operating conditions. This paper deals with fault detection in inverter‐fed EV using a dual‐tree complex wavelet transform (DTCWT) based squeeze net (SN) and optimized support vector machine (SVM). Due to the simple structure and high power density, most EV models on the market are equipped with induction motors. In the proposed work, the voltage, current, and speed signals are measured at different faulty conditions, and then the features are extracted through the DTCWT‐based SN. Extracted data are processed and classified through the sucker‐vulture optimization algorithm (SVOA) based SVM. In the proposed SVOA, the exploration phase of remora optimization algorithm is used for the exploitation phase of African vulture optimization algorithm (AVOA). Thus, the convergence speed of AVOA is improved. The proposed method is implemented in MATLAB/SIMULINK, and the results are used for different scenarios. The accuracy and F1‐score for the proposed methodology are attained as 99.92441% and 99.92441%. From the obtained results, it is clear that the proposed DTCWT‐based SN effectively detects the faults in the inverter.

Near-Infrared Off-Axis Cavity-Enhanced Optical Frequency Comb Spectroscopy for CO2/CO Dual-Gas Detection Assisted by Machine Learning.

Cavity-enhanced direct frequency comb spectroscopy (CE-DFCS) is widely used as a highly sensitive gas sensing technology in various gas detection fields. For the on-axis coupling incidence scheme, the detection accuracy and stability are seriously affected by the cavity-mode noise, and therefore, stable operation inevitably requires external electronic mode-locking and sweeping devices, substantially increasing system complexity. To address this issue, we propose off-axis cavity-enhanced optical frequency comb spectroscopy from both theoretical and experimental aspects, which is applied to the detection of single- and dual-gas of carbon monoxide (CO) and carbon dioxide (CO2) in the near-infrared. An erbium-doped fiber frequency comb with a repetition frequency of ∼41.709 MHz is coupled into a resonant cavity with a length of ∼360 mm in an off-axis manner, exciting numerous high-order modes to effectively suppress cavity-mode noise. The performance of multiple machine learning models is compared for the inversion of a single/dual gas concentration. A few absorbance spectra are collected to build a sample data set, which is then utilized for model training and learning. The results demonstrate that the Particle Swarm Optimization Support Vector Machine (PSO-SVM) model achieves the highest predictive accuracy for gas concentration and is ultimately applied to the detection system. Based on Allan deviation, the detection limit for CO in single-gas detection can reach 8.247 parts per million by volume (ppmv) by averaging 87 spectra. Meanwhile, for simultaneous CO2/CO measurement with highly overlapping absorbance spectra, the LoD can be reduced to 13.196 and 4.658 ppmv, respectively. The proposed optical gas sensing technique indicates the potential for the development of a field-deployable and intelligent sensor system capable of simultaneous detection of multiple gases.