The effect of principal component analysis on machine learning accuracy with high-dimensional spectral data

Howley, Tom; Madden, Michael G.; O’Connell, Marie-Louise; Ryder, Alan G.

doi:10.1016/j.knosys.2005.11.014

Cited by 111 publications

(49 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Both FE and FS have been found in many studies (Babaoğlu, Fındık, & Bayrak, 2010; Chandrashekar & Sahin, 2014; Howley, Madden, O'Connell, & Ryder, 2006) to improve classification accuracy, and it was therefore expected that predictions following these methods would be more accurate than predictions made on the full dataset. However, in the ecologically redundant Kwongan, these methods led to statistically significant classification improvements in only three scenarios, whereas in the nonredundant Woodland, they improved accuracies 12 times.…”

Section: Discussionmentioning

confidence: 99%

Impact of ecological redundancy on the performance of machine learning classifiers in vegetation mapping

MacIntyre

Niekerk

Dobrowolski

et al. 2018

Ecology and Evolution

View full text Add to dashboard Cite

Vegetation maps are models of the real vegetation patterns and are considered important tools in conservation and management planning. Maps created through traditional methods can be expensive and time‐consuming, thus, new more efficient approaches are needed. The prediction of vegetation patterns using machine learning shows promise, but many factors may impact on its performance. One important factor is the nature of the vegetation–environment relationship assessed and ecological redundancy. We used two datasets with known ecological redundancy levels (strength of the vegetation–environment relationship) to evaluate the performance of four machine learning (ML) classifiers (classification trees, random forests, support vector machines, and nearest neighbor). These models used climatic and soil variables as environmental predictors with pretreatment of the datasets (principal component analysis and feature selection) and involved three spatial scales. We show that the ML classifiers produced more reliable results in regions where the vegetation–environment relationship is stronger as opposed to regions characterized by redundant vegetation patterns. The pretreatment of datasets and reduction in prediction scale had a substantial influence on the predictive performance of the classifiers. The use of ML classifiers to create potential vegetation maps shows promise as a more efficient way of vegetation modeling. The difference in performance between areas with poorly versus well‐structured vegetation–environment relationships shows that some level of understanding of the ecology of the target region is required prior to their application. Even in areas with poorly structured vegetation–environment relationships, it is possible to improve classifier performance by either pretreating the dataset or reducing the spatial scale of the predictions.

show abstract

Section: Discussionmentioning

confidence: 99%

Impact of ecological redundancy on the performance of machine learning classifiers in vegetation mapping

MacIntyre

Niekerk

Dobrowolski

et al. 2018

Ecology and Evolution

View full text Add to dashboard Cite

show abstract

“…According to [58], for datasets with very low complexity (few PCs), the relevant information has been excluded during the process of PCA, which resulted in a lower classification accuracy for datasets after PCA. The PCA could give a higher classification accuracy to datasets with very high complexity (many PCs), where the dataset before PCA does not only have relevant information, but also contains noise [59]. With the presence of noise, the classifier over-fits the training data and thus does not generalize well.…”

Section: Supervised Machine Learning Analysis For Pattern Recognitionmentioning

confidence: 99%

Pollutant Recognition Based on Supervised Machine Learning for Indoor Air Quality Monitoring Systems

et al. 2017

View full text Add to dashboard Cite

Indoor air may be polluted by various types of pollutants which may come from cleaning products, construction activities, perfumes, cigarette smoke, water-damaged building materials and outdoor pollutants. Although these gases are usually safe for humans, they could be hazardous if their amount exceeded certain limits of exposure for human health. A sophisticated indoor air quality (IAQ) monitoring system which could classify the specific type of pollutants is very helpful. This study proposes an enhanced indoor air quality monitoring system (IAQMS) which could recognize the pollutants by utilizing supervised machine learning algorithms: multilayer perceptron (MLP), K-nearest neighbour (KNN) and linear discrimination analysis (LDA). Five sources of indoor air pollutants have been tested: ambient air, combustion activity, presence of chemicals, presence of fragrances and presence of food and beverages. The results showed that the three algorithms successfully classify the five sources of indoor air pollution (IAP) with a classification rate of up to 100 percent. An MLP classifier with a model structure of 9-3-5 has been chosen to be embedded into the IAQMS. The system has also been tested with all sources of IAP presented together. The result shows that the system is able to classify when single and two mixed sources are presented together. However, when more than two sources of IAP are presented at the same period, the system will classify the sources as 'unknown', because the system cannot recognize the input of the new pattern.

show abstract

“…Previous studies [116] have shown how the number of PCs used affects classification error among several different classification methods for high dimensional data such as…”

Section: Selection Of Pcs Based On Cumulative Percent Variancementioning

confidence: 99%

The use of microfluidics and dielectrophoresis for separation, concentration, and identification of bacteria

et al. 2016

View full text Add to dashboard Cite

Traditional bacterial analyses take one to two days under favorable conditions where the bulk of the time is spent waiting for bacteria to divide and grow until visual colonies can be observed for identification. In the case of bacteria with slow doubling times, this process can take weeks. This delay in analysis is unacceptable, especially in cases of life threatening diseases or emergencies. It is clear that in order to decrease the analysis time of the bacteria, the culturing and growth step must be circumvented. The goal of this research is to design, build, and test a device that could decrease the analysis time of bacteria using label-free methods of dielectrophoresis and Raman spectroscopy.Testing for device design was performed with clinical samples in mind, which consist of bacteria grown in a variety of environmental conditions (i.e. available food sources, growth stage, temperature, etc.) and accompanied by sample debris. Raman spectra of bacteria grown in varying media and metabolic stages were collected and analyzed. Results indicate that growth phase and media have an impact on Raman spectra iv and is distinguishable by linear discriminant analysis (LDA). Despite these spectral differences, it was found that LDA classification of closely related bacteria remains fairly high (90%) regardless of growth phase. Sample debris were also considered in device design and accommodated for by dielectrophoresis. Devices were built with the goal to isolate bacteria from a mixed sample and simultaneously acquire Raman spectra for identification.For this dissertation, a device was designed, built, and tested that incorporates dielectrophoresis for particle isolation and Raman spectroscopy for identification. The device was modeled in COMSOL to ensure that an appropriate electrical field gradient could be obtained to isolate bacteria from 5 µm diameter polystyrene spheres. The device was built and successfully trapped bacteria away from polystyrene spheres and Raman spectra of the bacteria were collected while trapped. These results indicate a clear potential for contactless dielectrophoresis-Raman devices to isolate and identify bacteria from sample debris, and thereby decrease the analysis time of bacteria. Typical bacterial analysis involves culturing and visualizing colonies on an array of agar plates. The growth patterns and colors among the array are used to identify the bacteria. For fast growing bacteria such as Escherichia coli, analysis will take one to two days. However, slow growing bacteria such as mycobacteria can take weeks to identify.In addition, there are some species of bacteria that are viable but nonculturable. This lengthy analysis time is unacceptable for life-threatening infections and emergency situations. It is clear that to decrease the analysis of the bacteria, the culturing and growth steps must be avoided. The goal of this research is to design, build, and test a device that could decrease the analysis time of bacteria.

show abstract

The effect of principal component analysis on machine learning accuracy with high-dimensional spectral data

Cited by 111 publications

References 12 publications

Impact of ecological redundancy on the performance of machine learning classifiers in vegetation mapping

Impact of ecological redundancy on the performance of machine learning classifiers in vegetation mapping

Pollutant Recognition Based on Supervised Machine Learning for Indoor Air Quality Monitoring Systems

The use of microfluidics and dielectrophoresis for separation, concentration, and identification of bacteria

Contact Info

Product

Resources

About