BEGIN:VCALENDAR VERSION:2.0 X-WR-CALNAME:EventsCalendar PRODID:-//hacksw/handcal//NONSGML v1.0//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/New_York LAST-MODIFIED:20240422T053451Z TZURL:https://www.tzurl.org/zoneinfo-outlook/America/New_York X-LIC-LOCATION:America/New_York BEGIN:DAYLIGHT TZNAME:EDT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZNAME:EST TZOFFSETFROM:-0400 TZOFFSETTO:-0500 DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT CATEGORIES:College of Engineering,Thesis/Dissertations DESCRIPTION:Advisor: Dr. Donghui Yan Committee Members: Dr. Yuchou Chang, Computer and Information Science Department, University of Massachusetts Dartmouth Dr. Long Jiao, Computer and Information Science Department, Univ ersity of Massachusetts Dartmouth Abstract: Cryptocurrency markets are non -stationary, making price forecasting inherently unreliable over time. Thi s study examines whether the choice of target variable has more impact on forecast stability than the choice of model architecture. Five models are evaluated across two target formulations: raw Bitcoin price and 1-hour per centage change. The models tested are Naive Forecast, ARIMA, LSTM, Bidirec tional LSTM, and GRU, each trained and assessed over a 27-day test window using live data collected at 5-minute intervals across a rolling six-week period. Drift detection using the Wasserstein distance confirmed that raw price exhibits significantly greater distributional shift than percentage change over the same timeframe. Models trained on raw price produced direc tional accuracy below 50% across all learned architectures, with visible d egradation over the test window. The Naive Forecast outperformed all learn ed models on both RMSE ($110.34) and MAE ($66.08). Models trained on perce ntage change maintained substantially higher accuracy: LSTM achieved 74.7% directional accuracy, while BiLSTM and GRU both reached 72.8%, with no co mparable decay observed. The results indicate that model decay in Bitcoin forecasting is driven primarily by data drift in the target variable rathe r than by limitations in the predicting architecture. When the target is s tationary, all models tested retain their accuracy across the full evaluat ion window.\nEvent page: https://www.umassd.edu/events/cms/data-science-ms -thesis-defense-by-adithi-madduluri.php X-ALT-DESC;FMTTYPE=text/html:

Advisor: Dr. Donghui Yan

\ n

Committee Members:

Dr. Yuchou Chang\, Computer and Information Science Department\, University of Massachusetts Dartmouth

Dr. Lo ng Jiao\, Computer and Information Science Department\, University of Mass achusetts Dartmouth

Abstract:

Cryptocurrency markets are n on-stationary\, making price forecasting inherently unreliable over time. This study examines whether the choice of target variable has more impact on forecast stability than the choice of model architecture. Five models a re evaluated across two target formulations: raw Bitcoin price and 1-hour percentage change. The models tested are Naive Forecast\, ARIMA\, LSTM\, B idirectional LSTM\, and GRU\, each trained and assessed over a 27-day test window using live data collected at 5-minute intervals across a rolling s ix-week period. Drift detection using the Wasserstein distance confirmed t hat raw price exhibits significantly greater distributional shift than per centage change over the same timeframe. Models trained on raw price produc ed directional accuracy below 50% across all learned architectures\, with visible degradation over the test window. The Naive Forecast outperformed all learned models on both RMSE ($110.34) and MAE ($66.08). Models trained on percentage change maintained substantially higher accuracy: LSTM achie ved 74.7% directional accuracy\, while BiLSTM and GRU both reached 72.8%\, with no comparable decay observed. The results indicate that model decay in Bitcoin forecasting is driven primarily by data drift in the target var iable rather than by limitations in the predicting architecture. When the target is stationary\, all models tested retain their accuracy across the full evaluation window.

Event page: https:/ /www.umassd.edu/events/cms/data-science-ms-thesis-defense-by-adithi-maddul uri.php

DTSTAMP:20260408T165644 DTSTART;TZID=America/New_York:20260429T103000 DTEND;TZID=America/New_York:20260429T113000 LOCATION:Virtual SUMMARY;LANGUAGE=en-us:Data Science MS Thesis Defense by Adithi Madduluri UID:ab7899b3e0a57697b2ec538385b0420d@www.umassd.edu END:VEVENT END:VCALENDAR