Dynamic Modeling and Forecasting of Time-evolving Data Streams

Matsubara, Yasuko; Sakurai, Yasushi

doi:10.1145/3292500.3330947

Cited by 15 publications

(12 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Mining and modeling time-series data is a building block of many analytical and predictive tasks, such as pattern discovery [ 11 , 12 ], disaggregation [ 13 ], and forecasting [ 2 , 3 , 14 , 15 ], in a variety of fields, including social media [ 16 , 17 ], web [ 14 ], and medical science [ 18 ]. Especially, ordinary differential equations (ODEs) have attracted much attention, due to its simplicity and expressiveness, and several studies focus on learning ODEs from data [ 19 – 22 ].…”

Section: Related Workmentioning

confidence: 99%

“…While these models are intuitive and simple, they often have limited expressiveness, failing to capture epidemic dynamics accurately. On the other hand, data-driven models [ 2 , 3 ] aim to model and forecast co-evolving time-series data using ODEs, without relying on human knowledge. They employ latent variables and non-linear differential equations to capture complicated temporal dynamics.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Simple epidemic models with segmentation can be better than complex ones

2022

View full text Add to dashboard Cite

Given a sequence of epidemic events, can a single epidemic model capture its dynamics during the entire period? How should we divide the sequence into segments to better capture the dynamics? Throughout human history, infectious diseases (e.g., the Black Death and COVID-19) have been serious threats. Consequently, understanding and forecasting the evolving patterns of epidemic events are critical for prevention and decision making. To this end, epidemic models based on ordinary differential equations (ODEs), which effectively describe dynamic systems in many fields, have been employed. However, a single epidemic model is not enough to capture long-term dynamics of epidemic events especially when the dynamics heavily depend on external factors (e.g., lockdown and the capability to perform tests). In this work, we demonstrate that properly dividing the event sequence regarding COVID-19 (specifically, the numbers of active cases, recoveries, and deaths) into multiple segments and fitting a simple epidemic model to each segment leads to a better fit with fewer parameters than fitting a complex model to the entire sequence. Moreover, we propose a methodology for balancing the number of segments and the complexity of epidemic models, based on the Minimum Description Length principle. Our methodology is (a) Automatic: not requiring any user-defined parameters, (b) Model-agnostic: applicable to any ODE-based epidemic models, and (c) Effective: effectively describing and forecasting the spread of COVID-19 in 70 countries.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Simple epidemic models with segmentation can be better than complex ones

2022

View full text Add to dashboard Cite

show abstract

“…In the meantime, and, as the M4 competition was running, many new methods were introduced. Novel approaches were presented ranging from linear differential equations [27], regularized regression [28], and of course deep learning approaches like Deep Clustering [29], Recurrent Neural Networks [30], Deep Ensembles [31] , Attention Networks [32], and Convolutional Networks [33]. All these methods were applied in multiple application domains ranging from weather forecasting to pandemics.…”

Section: Conclusion and The Future Of Forecastingmentioning

confidence: 99%

A Multi-Factor Analysis of Forecasting Methods: A Study on the M4 Competition

Agathangelou

Trihinas

Katakis

2020

Data

View full text Add to dashboard Cite

As forecasting becomes more and more appreciated in situations and activities of everyday life that involve prediction and risk assessment, more methods and solutions make their appearance in this exciting arena of uncertainty. However, less is known about what makes a promising or a poor forecast. In this article, we provide a multi-factor analysis on the forecasting methods that participated and stood out in the M4 competition, by focusing on Error (predictive performance), Correlation (among different methods), and Complexity (computational performance). The main goal of this study is to recognize the key elements of the contemporary forecasting methods, reveal what made them excel in the M4 competition, and eventually provide insights towards better understanding the forecasting task.

show abstract

“…Researchers have focused on the advantages of ML models to succeed regardless of the size of data (Matsubara and Sakurai, 2019), or on the combinations of ML models that capture across-multiple-series features with individual-series approaches (Deshpande and Sarawagi, 2019). Techniques for time series selection have also grown in complexity with decision trees, including geo-localization, (Chatzigeorgakidis et al, 2019).…”

Section: Machine Learning In Sequential Data Predictionmentioning

confidence: 99%

“…In both cases, they have a ground-truth to be used to assess the model. In the (Matsubara and Sakurai, 2019) and (Hatt and Feuerriegel, 2020) cases, there were also known ground-truth to train the data. By contrast, in this work there is no enough data to build the ground-truth to drive the experiments, data from early stages of the running system is used to build the initial "ground truth".…”

Section: Machine Learning In Sequential Data Predictionmentioning

confidence: 99%

A Hybrid Learning Approach for Modelling and Analyzing Clickstream Data from Learning Management Systems

Hechavarria¹

View full text Add to dashboard Cite

In this research, a hybrid approach combining a Hidden Markov Model (HMM) with a Long Short-Term Memory (LSTM) recurrent neural network (RNN) is introduced to model realtime online feedback to students when completing academic activities using online Learning Management Systems (LMS). The solution provided is a Smart Classifier which unravels, and processes hidden patterns in the data to train appropriate metrics to raise flags indicating outlier student behavior based on historical data from previous and ongoing sessions. This work introduces an approach that facilitates modifications of the attention mechanism in Transformer models. Using this approach, the predictor module of the proposed solution is improved. The key element of this improvement is to use a Bayesian Graph Network (BGN) coupled to a Transformer. As a novelty, this method provides a systematic customization of the attention mechanism in Transformer models that can be applied to a range of problems involving clickstream data.iii AcknowledgementsBefore anything else, I wish to express my deepest gratitude to my supervisor, Dr. M.Omair Shafiq for his unwavering support and belief in me. This thesis and the research behind it have been continuously guided by his enthusiasm, knowledge, research and teaching excellence in the field. From the early stages of my transition into the program and subsequent course work, to the completion of this thesis under the remote work and studying conditions imposed on us all by the pandemic, his remarkably constructive and effective feedback, insightful comments, and encouragement have been crucial for my success at every step of this journey. He has taught me very valuable lessons in a demanding field, always in the most professional, and yet warm and friendly, manner. This work would not have been possible without the financial support of the Carleton University COVID-19 Rapid Response Research Grant, awarded to Dr. M. Omair Shafiq in 2020; and the funding opportunity from the School of Information Technology to undertake my studies. I would like to express my sincere thanks to Dr. Audrey Girouard, Associate Director, Graduate Studies for her overall guidance, and to the faculty and staff of the School of Information Technology for providing students with a stable and highly successful academic program amidst the unprecedented COVID-19 pandemic. I would like to extend my special thanks to my professors in the program at the School of Information Technology and the School of Computer Science, especially Dr. Olga Baysal, Dr. Elio Velazquez, and PhD. Candidate Mohamed Abdelazez, for their valuable work in shaping some of the key methods and techniques applied in this research. I would like iv to express my gratitude to Dr. David Sprague for his kind help and support allowing me to contribute to his Teaching Assistance team, gaining valuable and practical knowledge while helping students through the last two years. I would like to thank all members of the Administrative Staff at the School of Information Technology, a...

show abstract

Dynamic Modeling and Forecasting of Time-evolving Data Streams

Cited by 15 publications

References 29 publications

Simple epidemic models with segmentation can be better than complex ones

Simple epidemic models with segmentation can be better than complex ones

A Multi-Factor Analysis of Forecasting Methods: A Study on the M4 Competition

A Hybrid Learning Approach for Modelling and Analyzing Clickstream Data from Learning Management Systems

Contact Info

Product

Resources

About