Sex Determination of Three-Dimensional Skull Based on Improved Backpropagation Neural Network

Area of Science:

• Forensic Anthropology
• Biometrics
• Artificial Intelligence

Background:

• Skeletal sex determination is crucial in forensic anthropology.
• Traditional methods rely on subjective analysis of sexually dimorphic features.
• Objective and accurate methods are needed for reliable sex assessment.

Purpose of the Study:

• To develop an improved backpropagation neural network (BPNN) for accurate sex determination from skull morphology.
• To enhance the BPNN with momentum, regularization, and Adaboost for improved performance.
• To compare the BPNN model's accuracy against traditional forensic anthropology methods.

Main Methods:

• Utilized a dataset of 267 skulls (153 female, 114 male).
• Employed computer-aided measurements of six skull characteristics as input features.
• Implemented and experimented with two BPNN architectures: [6; 6; 2] and [6; 12; 2].
• Optimized BPNN parameters, including learning rate (η) and momentum (α).

Main Results:

• The [6; 12; 2] BPNN architecture demonstrated superior average accuracy.
• Optimal classification accuracy was achieved with η=0.5 and α=0.9.
• The model reached 97.232% accuracy in training and 96.764% in testing, with low Mean Squared Error (MSE).

Conclusions:

• The proposed improved BPNN offers enhanced learning ability and faster convergence for skeletal sex determination.
• This AI-driven approach significantly outperforms traditional methods in classification accuracy.
• The BPNN provides a more objective and reliable tool for forensic anthropologists.