Trade data is typically time-series data. However, relying on a single time-series forecasting algorithm may not provide the most accurate predictions. To improve forecasting performance, our prediction module integrates time-series forecasting algorithms with machine learning techniques.
First, after obtaining the historical dataset for a specific forecast condition, we perform trend, seasonal, and cyclical analyses on the data. This helps to identify the trend patterns, seasonal patterns, and cycles, and to define the range of parameters to be included in the model. (The historical dataset refers to the data from the GTI International Trade Statistics Database.)
Next, we use Optuna, a tool for hyperparameter optimization, to fine-tune the integrated model for optimal performance. We evaluate the model's performance using Root Mean Square Error (RMSE) as the metric. To ensure more accurate predictions, our training set includes at least 60 months of historical data, with a validation set consisting of at least 5 months of data.
Finally, the trained model is used to predict data for the time range selected by the user.
Covers 250 Countries/Economies in the GTI Global Trade Statistics Database
Sorry, there isn't enough historical data for this condition. Please try different condition.
全球贸易与产业经济全景式俯瞰
Copyright © 2024 Dalian Infobank Co.,Ltd. All Rights Reserved。
请前往PC端查询!
Due to a scheduled power outage in our business park, our website will be unavailable from 8:00 AM to 8:00 PM Beijing Time on Sunday, October 20th, 2024. We apologize for any inconvenience this may cause.
Do not remind me