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. Patrick Kreidl

Second Advisor

Dr. Karthikeyan Umapathy

Third Advisor

Dr. Mona Nasseri


Human infertility rates are on the rise -- studies have shown that one in every four women have trouble conceiving and every year those rates are rising [Halleran22]. This trend motivates procedures such as in-vitro fertilization (IVF) and intrauterine insemination (IUI) to produce a pregnancy. Such procedures are not only intricate and expensive but also carry potential unintended consequences for women's physical and mental well-being. A more profound comprehension of the menstrual cycle and its multiple phases, particularly detecting the ovulation period itself, promotes optimized treatment outcomes and reduces associated risks. To facilitate ovulation detection, new sensing technologies built into noninvasive wearable devices are actively under study [Polanski22]. These wearable devices measure and collect different physiological signals from which features can be extracted that are hypothesized to be helpful towards predicting when ovulation is more likely to occur.

This thesis strives to classify the subsets of days in which ovulation occurs based upon a specific set of thirteen features per day, all extracted by domain experts from human subject data collected via the Empatica E4 wristband [Sides23]. Building on the prior work, we have developed classifiers for the provided per-day feature data are developed via a well-established machine learning algorithm, called Random Forest, and their performance is evaluated using standard classification metrics. The findings reveal that a classification accuracy of up to 74% is attainable for the implementation focused solely on intraday feature correlations, while near-perfect accuracy is attained by the implementation also incorporating interday feature correlations. Inspection of the predictors themselves also suggests the subset of per-day features that most distinguishes ovulation from non-ovulation, which is observed to remain consistent in the intraday and interday implementations. Noting that our study is entirely in the context of offline classification, by virtue of the per-day features extracted from post-processing of wearable device data, also discussed are unresolved challenges associated with attaining comparable accuracies for real-time ovulation detections and tracking objectives.

Available for download on Friday, December 12, 2025