Hyperparameter Optimization of Machine Learning Models for Monitoring Menstrual Mental Health

Monitoring and tracking mood disorders associated with Menstruation are becoming crucial to avoid impairment in the mental health of a woman, prevent disruption in family life and relationships, and allow women to maintain a proper balance in their work life. Every sentiment linked during the menstruation cycle can be identified for improving fecundability and health management. This study aims to analyze the feelings of women during Premenstrual dysphoric disorder (PMDD) and the actual menstrual cycle using text-mining methods. We used sentimental analysis in a Reddit dataset to identify positive and negative emotions in women. In the proposed framework, public sentiments were first processed and data was balanced using Synthetic Minority Over-sampling Technique (SMOTE) followed by classification using different machine learning frameworks like Support Vector Machine (SVM), Decision Tree Classifier, Logistic Regression, Random Forest, AdaBoost Classifier and Boosting Classifier. The findings of this study exhibit that women evoke both positive and negative sentiments. However, compared to other feelings the negative emotion is at its inception. The result shows that Random Forest with tuned hyperparameters gives us the best accuracy of 94% when compared to other baseline and ensemble machine learning models with different feature extraction techniques.