Principal Component Analysis Preprocessing Research Articles

Utilizing principal component analysis to enhance machine learning in bankruptcy prediction: A comparative investigation

In the global economic environment, bankruptcy prediction is essential for managing financial risks. With the advancement of machine learning techniques, accurate prediction using these methods has become a research focus. This study explores the efficiency and accuracy of improved machine learning models for bankruptcy prediction through principal component analysis (PCA). By utilizing a dataset from the banking industry in Taiwan, this paper compares the performance of PCA with and without processed data using logistic regression modeling. The research methodology includes data preprocessing, PCA downscaling, and subsequent model training and testing. The key research question is whether PCA preprocessing can significantly improve the operational efficiency and predictive accuracy of the model. It is found that the model with PCA outperforms the model without PCA in terms of processing time and accuracy. This suggests that PCA can effectively improve the performance of bankruptcy prediction models and provide a more effective tool for financial risk management. These findings provide useful insights into other areas of financial analysis using machine learning and support the value of applying PCA in predictive modeling.

Detection and diagnosis of bacterial pathogens in urine using laser-induced breakdown spectroscopy

The presence of bacterial cells from three species has been detected in clinical specimens of human urine using laser-induced breakdown spectroscopy (LIBS) by using a partial least squares discriminant analysis (PLS-DA) of 360 spectra obtained from 12 specimens of infected urine and 239 spectra obtained from eight specimens of sterile urine. Nominally sterile urine specimens obtained from four patients at a local hospital after being screened negative for the presence of bacterial pathogens were spiked with known aliquots of Escherichia coli, Staphylococcus aureus, and Enterobacter cloacae to simulate clinical urinary tract infections. Fifteen emission line intensities measured from the LIBS spectra and 92 ratios of those line intensities were used as 107 independent variables in the PLS-DA for discrimination between bacteria-containing specimens and sterile specimens. The PLS-DA models possessed a 98.3% sensitivity and a 97.9% specificity for the detection of pathogenic cells in urine when single-shot LIBS spectra were tested. To increase the signal to noise ratio, thirty spectra acquired from a single specimen were also averaged together and the averaged spectra were used to construct a model. When each averaged spectrum was withheld from the model individually for testing, the diagnostic test possessed a 100% sensitivity and a 100% specificity for the detection of bacterial cells in urine, although the number of test spectra was necessarily reduced.The entire LIBS spectrum from 200 nm – 590 nm was input into an artificial neural network analysis with principal component analysis pre-processing (PCA-ANN) to diagnose the bacterial species once detected. This PCA-ANN test possessed an overall sensitivity of 97.2%, an overall specificity of 98.6%, and an overall classification accuracy of 97.9% when using 80% of the data to build a model and withholding 20% for cross-validation testing. The PCA-ANN was also performed on each of the 12 bacteria-containing filters individually, using the other 11 filters to build the model. The average sensitivity of this test, calculated by averaging the sensitivities measured for each of the three bacterial species, was 70.9% and the average specificity was 85.5%. Based on these results, the average classification accuracy for the test when used to discriminate between the three microorganisms was 80.6%. These results indicate the potential usefulness of LIBS for rapidly detecting and possibly diagnosing urinary tract infections in a clinical setting.

Detection and diagnosis of bacterial pathogens in blood using laser-induced breakdown spectroscopy

The ability to rapidly and accurately detect and identify pathogenic bacteria in clinically-obtained blood specimens with laser-induced breakdown spectroscopy (LIBS) is evaluated. Samples of blood obtained from five patients in a local hospital were confirmed to be negative for the presence of bacteria by the pathology department and were then tested with LIBS. Specimens of blood were tested as obtained from the hospital with no other alteration as control samples and were also intentionally spiked with known aliquots of Escherichia coli, Staphylococcus aureus, Enterobacter cloacae, and Pseudomonas aeruginosa to simulate blood infections. LIBS spectra were acquired from blood deposited on nitrocellulose filters. The intensities of 15 emission lines measured in the spectra and 92 ratios of those line intensities were used as 107 independent variables in a partial least squares discriminant analysis (PLS-DA) to discriminate between sterile control samples and those spiked with bacteria. In addition, the entire LIBS spectrum from 200 nm – 590 nm was input into an artificial neural network analysis with principal component analysis pre-processing (PCA-ANN) to diagnose the bacterial species once detected.The PLS-DA test possessed a 96.3% sensitivity and a 98.6% specificity for the detection of pathogenic bacteria in blood when 776 spectra from 26 filters were tested by removing one entire filter at a time from the model and testing each spectrum individually. When all the spectra obtained from a filter were averaged to enhance the signal to noise of the spectrum, 19 of 19 filters of infected blood tested positive and 7 of 7 filters with sterile blood tested negative, yielding 100% sensitivity and 100% specificity. The PCA-ANN test performed on the entire LIBS spectrum possessed a 100% sensitivity and 100% specificity when using 80% of the data to build a model and withholding 20% for cross-validation testing. The same PCA-ANN performed on each of the 19 filters individually, using the other 18 filters to build the model, possessed an average sensitivity of 85.5%, an average specificity of 95.0%, and a classification accuracy of 92.5%. These results indicate the potential usefulness of LIBS for detecting and diagnosing blood infections in a clinical setting.

Eliminating polarization leakage effect for neutral hydrogen intensity mapping with deep learning

ABSTRACT The neutral hydrogen (H i) intensity mapping (IM) survey is regarded as a promising approach for cosmic large-scale structure studies. A major issue for the H i IM survey is to remove the bright foreground contamination. A key to successfully removing the bright foreground is to well control or eliminate the instrumental effects. In this work, we consider the instrumental effects of polarization leakage and use the U-Net approach, a deep learning-based foreground removal technique, to eliminate the polarization leakage effect. The thermal noise is assumed to be a subdominant factor compared with the polarization leakage for future H i IM surveys and ignored in this analysis. In this method, the principal component analysis (PCA) foreground subtraction is used as a pre-processing step for the U-Net foreground subtraction. Our results show that the additional U-Net processing could either remove the foreground residual after the conservative PCA subtraction or compensate for the signal loss caused by the aggressive PCA pre-processing. Finally, we test the robustness of the U-Net foreground subtraction technique and show that it is still reliable in the case of existing constraint error on H i fluctuation amplitude.

Fast characterization of biomass pyrolysis oil via combination of ATR-FTIR and machine learning models

This study proposed a fast characterization method of bio-oil via the combination of attenuated total reflection flourier transformed infrared spectroscopy (ATR-FTIR) and machine learning models. The input to the model is high-dimensional infrared spectral data. Unsaturated concentration, effective hydrocarbon ratio, low calorific value, C content, H content, and O content are all relevant bio-oil indicators. The model parameters were optimized based on prediction accuracy and correlation coefficient. By comparing the sole support vector regression (SVR) model versus principal component analysis (PCA) preprocessed SVR model, the results showed that PCA preprocessing can significantly improve the overall performance of SVR model towards prediction of bio-oil characteristics. Under optimal parameters, the predicted accuracies for unsaturated concentration, effective hydrocarbon ratio, low calorific value, C content, H content, and O content reached 91.98%, 97.44%, 99.50%, 98.65%, 98.56%, and 97.88%, respectively. The correlation coefficient of sole SVR model was 0.3, and the correlation coefficient of PCA preprocessed SVR model was 0.9. Furthermore, the characteristic peaks of the infrared spectra at the optimal PC were analyzed, and PC6 and PC7 were found to have the most influence on the predicting performance.

High-Agreement Uncorrelated Secret Key Generation Based on Principal Component Analysis Preprocessing

Random and high-agreement secret key generation from noisy wideband channels is challenging due to the autocorrelation inside the channel samples and compromised cross correlation between channel measurements of two keying parties. This paper studies the signal preprocessing algorithms to establish high-agreement uncorrelated secret key in the presence of channel independent eavesdroppers. We first propose a general mathematical model for various preprocessing schemes, including principal component analysis (PCA), discrete cosine transform (DCT) and wavelet transform (WT). Among preprocessing schemes, PCA is proved to achieve the optimal secret key rate. Next, PCA with common eigenvector has been found to outperform PCA with private eigenvector in terms of an overall consideration of key agreement, information leakage, and computational expense. Then, we propose a system level design of key generation, including quantization, information reconciliation, and privacy amplification. Numerical results verify that the key generation enhanced by PCA with common eigenvector can achieve secret key with high key generation rate, low key error rate, and good randomness.

A Pilot Study on Brain Source Localization and Connectivity Analysis with MEG Responses to Unilateral Tactile Stimuli in Healthy Children Using Normalized Principal Component Analysis

Study on the interactions between brain regions during electrophysiological measurements provides deep insights into the mechanisms of brain information processing. However, the strong spatial correlations from background spontaneous activity might mask the local evoked activity and lead to spurious connectivity results in connectivity analysis for event-related EEG and MEG. In this paper, we applied normalized Principal Component Analysis (PCA) preprocessing, Minimum Norm Estimation (MNE) and Granger causal connectivity analysis to identify stimulus-elicited neural sources and measure the effective connectivity among the sources for left somatosensory MEG responses of five healthy subjects. After normalized PCA preprocessing our results demonstrated increased coherences between channels with somatosensory evoked response at major physiological frequency bands, including Alpha and Beta band for somatosensory evoked response. The sources were located at contralateral primary somatosensory cortex, contralateral secondary somatosensory cortex and cingulate cortex. The information flow was processed through a complex pattern of both feedforward and feedback interactions between the sources. Compare with the results without normalized PCA preprocessing, our results are consistent with established connections between somatosensory regions and previous source modeling studies, which provides validation of the proposed approach.

Cancer Discrimination Using Fourier Transform Near-Infrared Spectroscopy with Chemometric Models

Near-infrared (NIR) spectroscopy technique offers many potential advantages as tool for biomedical analysis since it enables the subtle biochemical signatures related to pathology to be detected and extracted. In conjunction with advanced chemometrics, NIR spectroscopy opens the possibility of their use in cancer diagnosis. The study focuses on the application of near-infrared (NIR) spectroscopy and classification models for discriminating colorectal cancer. A total of 107 surgical specimens and a corresponding NIR diffuse reflection spectral dataset were prepared. Three preprocessing methods were attempted and least-squares support vector machine (LS-SVM) was used to build a classification model. The hybrid preprocessing of first derivative and principal component analysis (PCA) resulted in the best LS-SVM model with the sensitivity and specificity of 0.96 and 0.96 for the training and 0.94 and 0.96 for test sets, respectively. The similarity performance on both subsets indicated that overfitting did not occur, assuring the robustness and reliability of the developed LS-SVM model. The area of receiver operating characteristic (ROC) curve was 0.99, demonstrating once again the high prediction power of the model. The result confirms the applicability of the combination of NIR spectroscopy, LS-SVM, PCA, and first derivative preprocessing for cancer diagnosis.

Class imbalance and its effect on PCA preprocessing

The performance of classification models is prone to the class imbalance problem, which occurs when one class of data severely outnumbers the other class. Solutions were proposed both at data level and algorithm level to improve the model performance in this phenomenon. Among all, resampling solutions which preprocess the class information at data level, are successfully applied in solving many real-world class imbalance problems. However, principal component analysis (PCA) is one of the prominent preprocessing solution to improve the classifier performance. PCA comprises new subspace from original attributes by maximising the global variance. This work explored the effect of class imbalance on the reduced subspace generated by the principal component analysis (PCA) for two-class classification problem. Initially the effect of class imbalance over PCA preprocessing is studied on synthetic datasets. Obtained results are further validated over ten real-world datasets. This study reveals two major findings: 1) whenever the angular separation between the respective principal axes of majority and minority classes is large then the data imbalance clearly affects the minority class prediction accuracy as well as the minority class data reconstruction from the principal eigen vectors of the combined datasets; 2) balancing the class distribution is crucial to ameliorates the classifier's performance than PCA preprocessing.

Classification of Gases/Odors Using Dynamic Responses of Thick Film Gas Sensor Array

This paper proposes a new method for classification of gases/odors called average slope multiplication (ASM) using dynamic characteristics of thick film gas sensor array. The instantaneous values of the extracted dynamic response/recovery plots for various test gases viz., LPG, CCl4, CO, and C3H7OH were correlated to its neighboring response plots by the use of proposed ASM technique. It has been demonstrated that the proposed method offers excellent results for classification of gases/odors using the dynamic responses of thick film gas sensor array. Principal component analysis (PCA) has been further used for data preprocessing and dimensionality reduction. The extracted raw data, the ASM transformed data, and PCA preprocessed ASM data were trained and tested using the back-propagation neural network (BPNN). The results thus obtained have been studied and presented here. Cross validation scheme was adopted for all analysis. The BPNN trained and tested with raw data showed 86% classification accuracy, whereas the raw data after PCA preprocessing showed 90% classification. The ASM data showed 97% classification accuracy while ASM data with PCA preprocessing showed the best results giving 100% classification accuracy with duly trained BPNN. We therefore report that superior identification of gases/odors can be obtained using the dynamic response of gas sensor array with the proposed ASM method.

Facilitating Neuronal Connectivity Analysis of Evoked Responses by Exposing Local Activity with Principal Component Analysis Preprocessing: Simulation of Evoked MEG

When connectivity analysis is carried out for event related EEG and MEG, the presence of strong spatial correlations from spontaneous activity in background may mask the local neuronal evoked activity and lead to spurious connections. In this paper, we hypothesized PCA decomposition could be used to diminish the background activity and further improve the performance of connectivity analysis in event related experiments. The idea was tested using simulation, where we found that for the 306-channel Elekta Neuromag system, the first 4 PCs represent the dominant background activity, and the source connectivity pattern after preprocessing is consistent with the true connectivity pattern designed in the simulation. Improving signal to noise of the evoked responses by discarding the first few PCs demonstrates increased coherences at major physiological frequency bands when removing the first few PCs. Furthermore, the evoked information was maintained after PCA preprocessing. In conclusion, it is demonstrated that the first few PCs represent background activity, and PCA decomposition can be employed to remove it to expose the evoked activity for the channels under investigation. Therefore, PCA can be applied as a preprocessing approach to improve neuronal connectivity analysis for event related data.

Dimensionality Reduction and Channel Selection of Motor Imagery Electroencephalographic Data

The performance of spatial filters based on independent components analysis (ICA) was evaluated by employing principal component analysis (PCA) preprocessing for dimensional reduction. The PCA preprocessing was not found to be a suitable method that could retain motor imagery information in a smaller set of components. In contrast, 6 ICA components selected on the basis of visual inspection performed comparably (61.9%) to the full range of 22 components (63.9%). An automated selection of ICA components based on a variance criterion was also carried out. Only 8 components chosen this way performed better (63.1%) than visually selected components. A similar analysis on the reduced set of electrodes over mid-central and centro-parietal regions of the brain revealed that common spatial patterns (CSPs) and Infomax were able to detect motor imagery activity with a satisfactory accuracy.

Electro-Mechanical Impedance-Based Wireless Structural Health Monitoring Using PCA-Data Compression and k-means Clustering Algorithms

This article presents a practical method for an electro-mechanical impedance-based wireless structural health monitoring (SHM), which incorporates the principal component analysis (PCA)-based data compression and k-means clustering-based pattern recognition. An on-board active sensor system, which consists of a miniaturized impedance measuring chip (AD5933) and a self-sensing macro-fiber composite (MFC) patch, is utilized as a next-generation toolkit of the electromechanical impedance-based SHM system. The PCA algorithm is applied to the raw impedance data obtained from the MFC patch to enhance a local data analysis-capability of the on-board active sensor system, maintaining the essential vibration characteristics and eliminating the unwanted noises through the data compression. Then, the root-mean square-deviation (RMSD)-based damage detection result using the PCA-compressed impedances is compared with the result obtained from the raw impedance data without the PCA preprocessing. Furthermore, the k-means clustering-based unsupervised pattern recognition, employing only two principal components, is implemented. The effectiveness of the proposed methods for a practical use of the electromechanical impedance-based wireless SHM is verified through an experimental study consisting of inspecting loose bolts in a bolt-jointed aluminum structure.