Scholarlink

<p>Time series analysis involves examining data collected at successive time intervals to uncover patterns, trends, and dependencies for predictive purposes across various domains. Challenges include managing non-stationarity, seasonal variations, and noisy data, with critical tasks like addressing missing timestamps or values that affect model accuracy. Specific applications, such as power consumption monitoring and stock market analysis, require adept handling of irregular patterns and volatility prediction amidst external influences. Selecting optimal techniques for handling missing data and choosing appropriate forecasting methods, each with unique hyperparameters, poses significant challenges. This paper introduces Preplysis, a framework designed to automate time series analysis, employ various cleaning techniques, tune forecasting model hyperparameters. It is a recommendation system for technique selection. This comprehensive approach aims to enhance efficiency, accuracy, and accessibility in time series analysis, for informed decision-making and predictive modeling. Preplysis achieved exceptional performance with an RMSE of 0.03 in univariate stock prediction, surpassing benchmarks set by hybrid LSTM and RNN models and in multivariate power consumption prediction, Preplysis recommended GRU with exponential smoothing, achieving an improved RMSE of 0.639 compared to an average of 0.83 across 12 models tested, highlighting its efficacy in optimizing model selection for superior predictive accuracy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Preplysis: automation of time series data preprocessing & analysis

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Mehak Usmani,
  • Zulfiqar Ali Memon

摘要

Time series analysis involves examining data collected at successive time intervals to uncover patterns, trends, and dependencies for predictive purposes across various domains. Challenges include managing non-stationarity, seasonal variations, and noisy data, with critical tasks like addressing missing timestamps or values that affect model accuracy. Specific applications, such as power consumption monitoring and stock market analysis, require adept handling of irregular patterns and volatility prediction amidst external influences. Selecting optimal techniques for handling missing data and choosing appropriate forecasting methods, each with unique hyperparameters, poses significant challenges. This paper introduces Preplysis, a framework designed to automate time series analysis, employ various cleaning techniques, tune forecasting model hyperparameters. It is a recommendation system for technique selection. This comprehensive approach aims to enhance efficiency, accuracy, and accessibility in time series analysis, for informed decision-making and predictive modeling. Preplysis achieved exceptional performance with an RMSE of 0.03 in univariate stock prediction, surpassing benchmarks set by hybrid LSTM and RNN models and in multivariate power consumption prediction, Preplysis recommended GRU with exponential smoothing, achieving an improved RMSE of 0.639 compared to an average of 0.83 across 12 models tested, highlighting its efficacy in optimizing model selection for superior predictive accuracy.