Adjusted binary classification (ABC) model in forensic science: An example on sex classification from handprint...

Area of Science:

• Forensic Science
• Biometrics
• Statistical Modeling

Background:

• Binary classification is crucial in forensic science for evidence analysis.
• Current methods often lack accuracy due to overlapping data.
• Accurate classification is vital for expert opinions and legal outcomes.

Purpose of the Study:

• To introduce a novel Adjusted Binary Classification (ABC) algorithm.
• To enhance classification accuracy and predictive values (PPV, NPV) to 95% in forensic applications.
• To address limitations of traditional binary classification methods in handling overlapping data.

Main Methods:

• Developed and applied the Adjusted Binary Classification (ABC) algorithm.
• Utilized linear discriminant analysis (LDA) with traditional single cut-off values and the ABC approach.
• Employed handprint measurements from 160 participants (80 males, 80 females) for sex classification models.
• Split data into training/cross-validation (70%) and testing (30%) sets.

Main Results:

• Traditional models achieved 78.2-93% accuracy, PPV, and NPV in cross-validation.
• ABC models reached 95% accuracy, PPV, and NPV in cross-validation, classifying 35.5-88.1% of specimens.
• In testing, ABC models achieved 97.3-100% accuracy and 95.4-100% PPV/NPV, applicable to 29.1-87.5% of specimens.
• The ABC approach demonstrated superior performance over traditional methods.

Conclusions:

• The ABC algorithm effectively adjusts classification models to meet predefined accuracy, PPV, and NPV targets.
• This method offers an efficient tool for binary classification in forensic settings.
• The ABC approach minimizes the likelihood of incorrect classifications in forensic evidence analysis.