Fibromyalgia (FM) is a chronic pain disorder marked by widespread physical symptoms and psychological comorbidities. While neuroimaging has dominated FM research, such methods are often impractical or inaccessible in routine clinical care. This study investigates whether self-reported features can effectively distinguish FM patients from healthy controls (HC) and identify clinically meaningful FM subtypes, using the Emo-Fibro dataset (N = 66; 33 FM, 33 HC). We trained and evaluated supervised machine learning models including Logistic Regression, Random Forest and Support Vector Machine on the Toronto Alexithymia Scale (TAS-20) and the Positive and Negative Affect Schedule (PANAS) and assessed feature importance using absolute coefficients, Gini importance, and permutation importance. The Random Forest model achieved the highest classification performance (Accuracy = 0.81, ROC-AUC = 0.82), indicating that these features alone can offer robust diagnostic insight. We then applied K-Means clustering to the FM group and identified two subtypes: high-distress versus lower-distress, characterized by emotional regulation, psychological burden, and affect. These findings suggest that patient-reported psychological data not only aid FM diagnosis but also reveal meaningful heterogeneity to guide personalized care. By focusing on accessible, self-reported measures, this study supports a practical and emotion-informed approach to FM research and management.