Time Series Analysis and Forecasting with Automated Machine Learning on a National ICD-10 Database

Area of Science:

• Medical informatics
• Public health
• Machine learning

Background:

• Machine learning (ML) applications in healthcare are expanding, but time series analysis remains underutilized due to its complexity.
• Automated machine learning (AutoTS) offers a solution to streamline complex time series analysis.

Purpose of the Study:

• To deploy an automated time series (AutoTS) machine learning approach for accurate disease forecasting.
• To identify the most effective ML models for predicting future disease incidence.

Main Methods:

• Utilized the nationwide ICD-10 (International Classification of Diseases, Tenth Revision) dataset of hospitalized patients in Romania (2008-2018).
• Generated time series datasets and applied AutoTS to forecast the ten deadliest diseases.
• Performed predictions at a NUTS 2 (Nomenclature of Territorial Units for Statistics) regional level for 2019-2020.

Main Results:

• Achieved highly accurate AutoTS predictions for the ten deadliest diseases.
• Generated disease forecast results at a regional level for 2019 and 2020.
• Demonstrated the feasibility of regional-level time series forecasting using a national ICD-10 dataset.

Conclusions:

• This study is the first to apply AutoTS for regional disease forecasting on a national ICD-10 dataset.
• AutoTS technology can enhance the efficiency of national health policy implementation.
• Automated time series forecasting provides valuable insights for public health decision-making.