Enhanced data point importance for efficient data splitting in classification models: application to olive oil...

Area of Science:

• Multivariate data analysis
• Chemometrics
• Food authenticity verification

Background:

• Selecting representative samples is vital for accurate one-class classification models.
• Traditional methods like Kennard-Stone (KS) are widely used but can be less efficient.
• Data Point Importance (DPI) offers a strategy for identifying structurally significant samples.

Purpose of the Study:

• To evaluate the Enhanced Data Point Importance (EDPI) method for sample selection in DD-SIMCA modeling.
• To compare EDPI's performance against the classical Kennard-Stone (KS) approach.
• To assess the efficiency and effectiveness of EDPI in reducing sample size and computational load.

Main Methods:

• Utilized a DPI-driven layered convex hull strategy to rank samples by structural significance.
• Applied EDPI for sample selection in DD-SIMCA modeling.
• Compared EDPI with the Kennard-Stone (KS) method using a dataset of pure olive oil samples and other oils for specificity testing.

Main Results:

• A DD-SIMCA model using 50 EDPI-selected samples achieved 100% sensitivity on the training set.
• EDPI-selected samples yielded comparable or superior specificity (100% for canola, hazelnut, sunflower oils; >96% for soya oil) versus KS-selected samples (60 samples).
• EDPI reduced the number of required samples and computational effort compared to the KS method.

Conclusions:

• EDPI is an efficient and effective strategy for representative sample selection in multivariate data analysis.
• EDPI offers practical advantages in chemometrics and food authenticity verification by optimizing sample selection.
• The EDPI method demonstrates potential for building robust classification models with fewer samples and reduced computational demands.