Paper Type

Master's Thesis


College of Computing, Engineering & Construction

Degree Name

Master of Science in Computer and Information Sciences (MS)



NACO controlled Corporate Body

University of North Florida. School of Computing

First Advisor

Dr. Xudong Liu

Second Advisor

Dr. Mona Nasseri

Third Advisor

Dr. Patrick Kreidl

Department Chair

Dr. Asai Asaithambi

College Dean

Dr. William Klostermeyer


This thesis aims to enhance understanding of women’s reproductive function by employing several machine learning algorithms to predict menstrual cycle phases based on physiological signals. Traditional self-reporting methods have proven error-prone, necessitating a data-driven alternative. Unlike existing approaches, this study utilizes a comprehensive set of multimodal features, including wrist skin temperature, electrodermal activity (EDA), interbeat interval (IBI), and heart rate variations recorded using a wristband. The inclusion of these features enables the prediction of next-day physiological data and menstrual cycle phases for both regular and irregular cycles. For next-day physiological data prediction, Autoregressive Integrated Moving Average (ARIMA) and Random Forest algorithms were applied to predict mean temperature, heart rate (HR), IBI, and EDA tonic values of all 15 subjects(43 cycles). ARIMA demonstrated strong performance in predicting temperature, HR, IBI, and EDA tonic, with RMSE values of 0.136 ± 0.098 (C°), 1.254 ± 0.412 (bpm), 0.015 ± 0.007 (s), and 0.171 ± 0.161 (µS), respectively. Simultaneously, Random Forest predicted the same signals with RMSE values of 0.133 ± 0.055 (C°), 1.348 ± 0.330 (bpm), 0.019 ± 0.007 (s), and 0.425 ± 0.215 (µS). The comparative analysis highlighted their complementary strengths, offering a robust framework for precise next-day predictions in daily subject cycles. Menstrual cycle phase prediction involved 33 cycles recorded from 11 ovulating subjects. Random Forest consistently achieved a robust accuracy of 91% when predicting three phases (P, O, L) using aggregated data from earlier cycles and testing on the last. Logistic Regression demonstrated commendable accuracy at 88%. Long Short-Term Memory (LSTM) models showed promise, achieving a superior accuracy of 91% with a comprehensive training set. These findings highlight the potential of both traditional machine learning and deep learning in menstrual cycle phase prediction.

Available for download on Monday, December 18, 2028