COMPARISON ANALYSIS OF FUZZY AI AND LLM IN ENHANCING WOMEN MENSTRUAL TRACKING APP

Abstract

This study compares fuzzy logic, traditional machine learning (LSTM and Random Forest), and large language models (LLMs) in predicting menstrual cycles, using user data in the Nona Woman app. Historical cycle logs, user demographics, and contextual variables (sleep, stress, symptoms) were pre-processed into JSON format. Model performance was evaluated using evaluation metrics (accuracy, precision, recall, F1 score, and mean absolute error (MAE)). Results show that advanced LLMs (Llama 3 70B, ChatGPT GPT-4o, and Gemma 12B) significantly outperform fuzzy logic models and fine-tuned LLM variants, which tended to hallucinate or overfit. In addition, LSTM and Random Forest models performed reliably for users with regular cycles but lagged behind the LLMs on predictive metrics. The study also considers business implications, where LLM inference costs (approximately $0.01–$0.12 per query) may limit large-scale deployment in commercial menstrual tracking apps. In conclusion, the findings suggest that LLMs offer superior menstrual cycle prediction, but practical deployment must balance model performance with operational cost and interpretability.