Novel Prediction Method Applied to Wound Age Estimation: Developing a Stacking Ensemble Model to Improve Predictive...

Area of Science:

• Forensic Science
• Biotechnology
• Computational Biology

Background:

• Accurate wound age determination is critical in forensic investigations.
• Current methods for estimating wound age lack sufficient accuracy and reliable biomarkers.
• There is a need for advanced techniques to improve forensic wound age analysis.

Purpose of the Study:

• To develop and validate machine learning models for predicting wound age.
• To evaluate the efficacy of various machine learning algorithms, including ensemble methods, for forensic applications.
• To identify optimal computational strategies for enhancing the accuracy of wound age estimation.

Main Methods:

• Collected tissue samples from rats at multiple time points post-injury (0-24 hours).
• Analyzed nine mRNA expression levels as potential biomarkers for wound age.
• Trained and compared five base machine learning algorithms (RF, SVM, MLP, GB, SGD) and 26 stacking ensemble models.
• Validated model performance using internal and external datasets.

Main Results:

• A stacking ensemble model combining Random Forest (RF), Support Vector Machine (SVM), and Multilayer Perceptron (MLP) achieved the highest accuracy (92.85%) and AUROC (0.93).
• The best stacking model demonstrated a low root-mean-square-error (RMSE) of 1.06 hours, indicating precise age prediction.
• The predictive performance of the ensemble models was consistently validated on an independent dataset.

Conclusions:

• Machine learning, particularly ensemble algorithms, shows significant potential for accurate wound age prediction in forensic science.
• The developed computational strategy offers a promising tool for improving forensic investigations.
• This approach may be adaptable for forecasting other forensic-related biological indicators.