Diagnosing anorexia based on partial least squares, back propagation neural network, and support vector machines

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare
• Nutritional Science

Background:

• Anorexia nervosa is a complex eating disorder with significant health consequences.
• Early diagnosis is crucial for effective treatment and improved patient outcomes.
• Current diagnostic methods can be invasive or lack predictive power.

Purpose of the Study:

• To develop and evaluate a novel predictive model for the early diagnosis of anorexia nervosa.
• To assess the efficacy of Support Vector Machine (SVM) compared to other machine learning algorithms.
• To identify key factors, including elemental concentrations and age, associated with anorexia nervosa.

Main Methods:

• Utilized a dataset of 90 cases with concentrations of six elements (Zn, Fe, Mg, Cu, Ca, Mn) and age.
• Developed a predictive model using Support Vector Machine (SVM) for early anorexia nervosa diagnosis.
• Compared SVM performance against Partial Least Squares (PLS) and Back-Propagation Neural Network (BPNN) classifiers.

Main Results:

• The SVM model achieved a high accuracy of 87% for the test set in diagnosing anorexia nervosa.
• SVM demonstrated superior performance compared to PLS (52% accuracy) and BPNN (65% accuracy).
• The developed models provided insights into factors correlated with anorexia nervosa.

Conclusions:

• Support Vector Machine (SVM) offers a promising, accurate, and non-invasive approach for early anorexia nervosa detection.
• Elemental analysis combined with age is a viable strategy for building predictive diagnostic models.
• This approach can aid clinicians in identifying at-risk individuals earlier, facilitating timely intervention.