Browsing by Subject "Model selection"

Now showing 1 - 3 of 3
  • Thumbnail Image
    Item
    AutoForecast: Automatic Time-Series Forecasting Model S
    (National Science Foundation, 2022) Abdallah, Mustafa; Rossi, Ryan; Mahadik, Kanak; Kim, Sungchul; Zhao, Handong; Bagchi, Saurabh; Electrical and Computer Engineering, Purdue School of Engineering and Technology
    In this work, we develop techniques for fast automatic selection of the best forecasting model for a new unseen time-series dataset, without having to first train (or evaluate) all the models on the new time-series data to select the best one. In particular, we develop a forecasting meta-learning approach called AutoForecast that allows for the quick inference of the best time-series forecasting model for an unseen dataset. Our approach learns both forecasting models performances over time horizon of same dataset and task similarity across different datasets. The experiments demonstrate the effectiveness of the approach over state-of-the-art (SOTA) single and ensemble methods and several SOTA meta-learners (adapted to our problem) in terms of selecting better forecasting models (i.e., 2X gain) for unseen tasks for univariate and multivariate testbeds.
  • Thumbnail Image
    Item
    Introducing COZIGAM: An R Package for Unconstrained and Constrained Zero-Inflated Generalized Additive Model Analysis
    (Foundation for Open Access Statistics, 2010-07-26) Liu, Hai; Chan, Kung-Sik; Medicine, School of Medicine
    Zero-inflation problem is very common in ecological studies as well as other areas. Nonparametric regression with zero-inflated data may be studied via the zero-inflated generalized additive model (ZIGAM), which assumes that the zero-inflated responses come from a probabilistic mixture of zero and a regular component whose distribution belongs to the 1-parameter exponential family. With the further assumption that the probability of non-zero-inflation is some monotonic function of the mean of the regular component, we propose the constrained zero-inflated generalized additive model (COZIGAM) for analyzingzero-inflated data. When the hypothesized constraint obtains, the new approach provides a unified framework for modeling zero-inflated data, which is more parsimonious and efficient than the unconstrained ZIGAM. We have developed an R package COZIGAM which contains functions that implement an iterative algorithm for fitting ZIGAMs and COZIGAMs to zero-inflated data basedon the penalized likelihood approach. Other functions included in the packageare useful for model prediction and model selection. We demonstrate the use ofthe COZIGAM package via some simulation studies and a real application.
  • Thumbnail Image
    Item
    System modeling reveals the molecular mechanisms of HSC cell cycle alteration mediated by Maff and Egr3 under leukemia
    (BMC, 2017-10-03) Li, Rudong; Wang, Yin; Cheng, Hui; Liu, Gang; Cheng, Tao; Liu, Yunlong; Liu, Lei; Medical and Molecular Genetics, School of Medicine
    Background Molecular mechanisms of the functional alteration of hematopoietic stem cells (HSCs) in leukemic environment attract intensive research interests. As known in previous researches, Maff and Egr3 are two important genes having opposite functions on cell cycle; however, they are both highly expressed in HSCs under leukemia. Hence, exploring the molecular mechanisms of how the genes act on cell cycle will help revealing the functional alteration of HSCs. Results We herein utilize the bioinformatic resources to computationally model the acting mechanisms of Maff and Egr3 on cell cycle. Using the data of functional experiments as reference, molecular acting mechanisms are optimally enumerated through model selection. The results are consolidated by evidences from gene sequence analysis, thus having enhanced the confidence of our pilot findings, which suggest that HSCs possibly undergo a “adaptation - suppression” process in response to the malignant environment of leukemia. Conclusion As a pilot research, our results may provide valuable insights for further experimental studies. Meanwhile, our research method combining computational modeling and data from functional experiments can be worthwhile for knowledge discovery; and it can be generalized and extended to other biological/biomedical studies. Electronic supplementary material The online version of this article (doi:10.1186/s12918-017-0467-4) contains supplementary material, which is available to authorized users.