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:College of EngineeringData Science Master's Thesis Defense "Ev aluating the Impact of Data Drift on Deep Learning Models for Bitcoin Pric e Forecasting" By Adithi Madduluri Advisor:  Dr. Donghui Yan, Mathemati cs, UMass Dartmouth Committee Members:Dr. Yuchou Chang, Computer and Info rmation Science Department, UMass DartmouthDr. Long Jiao, Computer and Inf ormation Science Department, UMass Dartmouth Thursday, April 29, 202610:3 0 am to 11:30 am Via Zoom: Please contact Adithi Madduluri (amadduluri@um assd.edu) or Dr. Yan (dyan@umassd.edu) for the zoom link and passcode  A bstract: Cryptocurrency markets are non-stationary, making price forecasti ng 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 are evaluated across two target formula tions: raw Bitcoin price and 1-hour percentage change. The models tested a re Naive Forecast, ARIMA, LSTM, Bidirectional LSTM, and GRU, each trained and assessed over a 27-day test window using live data collected at 5-minu te intervals across a rolling six-week period. Drift detection using the W asserstein distance confirmed that raw price exhibits significantly greate r distributional shift than percentage change over the same timeframe. Mod els trained on raw price produced directional accuracy below 50% across al l learned architectures, with visible degradation over the test window. Th e Naive Forecast outperformed all learned models on both RMSE ($110.34) an d MAE ($66.08). Models trained on percentage change maintained substantial ly higher accuracy: LSTM achieved 74.7% directional accuracy, while BiLSTM and GRU both reached 72.8%, with no comparable decay observed. The result s indicate that model decay in Bitcoin forecasting is driven primarily by data drift in the target variable rather than by limitations in the predic ting architecture. When the target is stationary, all models tested retain their accuracy across the full evaluation window. All Data Science and C omputer Science Graduate Students are encouraged to attend. For more info rmation, please contact Dr. Donghui Yan at dyan@umassd.edu.\nEvent page: h ttps://www.umassd.edu/events/cms/evaluating-the-impact-of-data-drift-on-de ep-learning-models-for-bitcoin-price-forecasting.php X-ALT-DESC;FMTTYPE=text/html:

College of Engineering
Dat a Science Master's Thesis Defense
 
"Evaluating the Impact of D ata Drift on Deep Learning Models for Bitcoin Price Forecasting"
 By Adithi Madduluri
 
Advisor:  
Dr. Donghui Yan\, Ma thematics\, UMass Dartmouth
 
Committee Members:
Dr. Yucho u Chang\, Computer and Information Science Department\, UMass Dartmouth
Dr. Long Jiao\, Computer and Information Science Department\, UMass Dar tmouth
 
Thursday\, April 29\, 2026
10:30 am to 11:30 am 
Via Zoom: Please contact Adithi Madduluri (amadduluri@umassd.e du) or Dr. Yan (dyan@umassd.edu) for the zoom link and passcode 
 < br />Abstract: Cryptocurrency markets are non-stationary\, making price fo recasting 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 are evaluated across two target formulations: raw Bitcoin price and 1-hour percentage change. The models t ested are Naive Forecast\, ARIMA\, LSTM\, Bidirectional LSTM\, and GRU\, e ach trained and assessed over a 27-day test window using live data collect ed at 5-minute intervals across a rolling six-week period. Drift detection using the Wasserstein distance confirmed that raw price exhibits signific antly greater distributional shift than percentage change over the same ti meframe. Models trained on raw price produced directional accuracy below 5 0% across all learned architectures\, with visible degradation over the te st window. The Naive Forecast outperformed all learned models on both RMSE ($110.34) and MAE ($66.08). Models trained on percentage change maintaine d substantially higher accuracy: LSTM achieved 74.7% directional accuracy\ , while BiLSTM and GRU both reached 72.8%\, with no comparable decay obser ved. The results indicate that model decay in Bitcoin forecasting is drive n primarily by data drift in the target variable rather than by limitation s in the predicting architecture. When the target is stationary\, all mode ls tested retain their accuracy across the full evaluation window.
 
All Data Science and Computer Science Graduate Students are encourag ed to attend.
 
For more information\, please contact Dr. Dongh ui Yan at dyan@umassd.edu.

Event page: https://www.umassd.edu/events/cms/evalu ating-the-impact-of-data-drift-on-deep-learning-models-for-bitcoin-price-f orecasting.php

DTSTAMP:20260410T163848 DTSTART;TZID=America/New_York:20260429T103000 DTEND;TZID=America/New_York:20260429T113000 LOCATION:via Zoom: please contact Adithi Madduluri (amadduluri@umassd.edu) or Dr. Yan (dyan@umassd.edu) for the zoom link and passcode SUMMARY;LANGUAGE=en-us:"Evaluating the Impact of Data Drift on Deep Learnin g Models for Bitcoin Price Forecasting" UID:93fc7d053397101c37313a7690143d43@www.umassd.edu END:VEVENT END:VCALENDAR