Scholarlink

Forecasting in time series remains a challenge despite numerous proposed models. There is a constant desire to improve the rates obtained, depending on the data available, especially in the context of quantity-limited data, which is the main challenge to overcome in this work. An approach based on sequential ensemble methods, with LSTM as a basic model called SEM-LSTM, is proposed to achieve this. The approach is based on the analysis of local autocorrelations, exploring temporal dependencies in specific segments of the series under study. A lag value d is deduced and taken as the sliding window size. The model is applied to Malaria data collected weekly in the Health Districts of the Adamawa Region of Cameroon to be assessed. The performance of SEM-LSTM is then compared with ARIMA and simple LSTM models using RMSE, MAE and R2 metrics. The results highlight the superiority of the SEM-LSTM model for predicting complex time series, such as Malaria cases, in the context of a limited quantity of data.

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

Time Series Forecasting Using the LSTM Model as a Sequential Ensemble Method

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Alioum Abdoulaye,
  • Apollinaire Batoure Bamana,
  • Richard Guiem,
  • Shadi Attala,
  • Kaladzavi Guidedi

摘要

Forecasting in time series remains a challenge despite numerous proposed models. There is a constant desire to improve the rates obtained, depending on the data available, especially in the context of quantity-limited data, which is the main challenge to overcome in this work. An approach based on sequential ensemble methods, with LSTM as a basic model called SEM-LSTM, is proposed to achieve this. The approach is based on the analysis of local autocorrelations, exploring temporal dependencies in specific segments of the series under study. A lag value d is deduced and taken as the sliding window size. The model is applied to Malaria data collected weekly in the Health Districts of the Adamawa Region of Cameroon to be assessed. The performance of SEM-LSTM is then compared with ARIMA and simple LSTM models using RMSE, MAE and R2 metrics. The results highlight the superiority of the SEM-LSTM model for predicting complex time series, such as Malaria cases, in the context of a limited quantity of data.