MLM

MLM improves data splitting for classification models in spectrochemical and biospectroscopic analyses by integrating a random-mutation factor into the Kennard-Stone algorithm to produce more balanced training and test sets.

Key Features:

• Random-mutation factor: Integrates stochastic mutations into the sample selection process to modify Kennard-Stone outcomes.
• Extension of Kennard-Stone (KS): Builds upon the KS algorithm for iterative selection of representative samples.
• Euclidean distance-based selection: Retains KS calculation of Euclidean distances between samples to ensure representative distribution of training and test sets.
• Balanced training/test sets: Aims to produce more balanced datasets to avoid unbalanced classification metrics.
• Improved sensitivity and specificity: Reported to improve sensitivity and specificity values relative to random selection (RS) and original KS.
• Benchmarking and evaluation: Evaluated against RS and KS across simulated scenarios and six real-world biospectroscopic applications.
• PCA-LDA evaluation: Performance assessed using principal component analysis linear discriminant analysis (PCA-LDA).
• Application domain: Targeted at spectrochemical analyses and biospectroscopic datasets for classification model development.

Scientific Applications:

• Spectrochemical analyses: Data splitting for classification models in spectrochemical studies.
• Biospectroscopic applications: Applied to six real-world biospectroscopic datasets for classifier development and validation.
• Biomedical classification models: Used to improve sensitivity and specificity in biomedical spectroscopic classification tasks.
• Simulation studies: Validated across simulated scenarios to benchmark performance against RS and KS.
• PCA-LDA-based evaluation: Suited for workflows that use principal component analysis linear discriminant analysis for model assessment.

Methodology:

Extends the Kennard-Stone (KS) Euclidean distance-based sample selection by incorporating a random-mutation factor for training/test split selection and assesses performance by comparing RS, KS, and MLM using principal component analysis linear discriminant analysis (PCA-LDA) on simulated scenarios and six real-world biospectroscopic datasets.