Docs Data Evaluation
Before using data for forecasting or model training, it is important to understand whether the data quality is sufficient. Data evaluation checks timestamps, missing values, series stability, and anomalies. It also scores data from two angles—integrity and forecastability—and then calculates an overall score. The results help you identify data issues and determine whether the data is suitable for modeling and forecasting analysis.
Start Evaluation
On the Data Evaluation page, select the dataset to evaluate and click Start Evaluation.
After the evaluation is complete, Latest Evaluation Results shows the evaluation time, four basic quality metrics, integrity, forecastability, and the overall score. Evaluation records retain the result of each evaluation. You can search by dataset name or refresh the list to view the latest status.
Basic Quality Metrics
The four basic quality metrics help identify specific issues in the data:
| Metric | Description | How to interpret it |
|---|---|---|
| Timestamp accuracy | Percentage of timestamps whose format and ordering meet requirements | Higher is better. A low value may indicate invalid formats, duplicate timestamps, or ordering issues. |
| Missing value rate | Percentage of missing points among all expected data points | Lower is better. A high value indicates gaps that may require filling or interpolation. |
| Stable series ratio | Percentage of relatively stable time series that are suitable for modeling | A higher value generally indicates greater overall stability. Also consider trends, cycles, and business context. |
| Anomaly rate | Percentage of outliers or abnormal fluctuations among all data points | Lower is better. A high value should be checked to determine whether it reflects a data quality issue or a real business event. |
Interpret all four metrics in business context. For example, anomalies caused by alerts or equipment startup and shutdown may be meaningful business signals and should not be removed solely because they increase the anomaly rate.
Scoring Results
Integrity
This score tells you whether the data has missing values and whether key fields are complete.
Sensor outages, network jitter, duplicate reports, missing fields, and similar issues can introduce gaps, duplicates, or incomplete structures in time-series data. The integrity score reflects how severe these issues are.
| Score | Meaning | Suggestion |
|---|---|---|
| 80–100 | Data integrity is good, key fields are complete, and modeling requirements are met | Safe to use directly |
| 60–79 | A small amount of data is missing, with limited impact on analysis results | Use after data governance |
| 0–59 | A large amount of data is missing, which may affect model training results | Prioritize data governance |
If the integrity score is low, forecasting or analysis results may be affected. Consider handling missing values, duplicate data, abnormal timestamps, or incomplete key fields before continuing with modeling.
Forecastability
This score tells you whether the series has learnable patterns.
Some data is naturally regular, such as electricity usage at fixed times of day or temperature that changes with seasons. Other data behaves more like random fluctuations. The forecastability score reflects how strong the patterns in the series are.
| Score | Meaning | Suggestion |
|---|---|---|
| 80–100 | Time-series features are stable and suitable for building forecasting models | Good candidate for modeling |
| 60–79 | The data has some regularity and can be used for forecasting analysis, but model performance may be affected by volatility | Try modeling and interpret results with business context |
| 0–59 | The data fluctuates significantly and is difficult to forecast | Review the data or forecasting strategy |
If the forecastability score is low, it does not necessarily mean the data is wrong. It may simply mean the series is highly volatile and hard to forecast. Use business context to decide whether to introduce additional variables or lower expectations for forecasting accuracy.
Overall Score
This score tells you whether the dataset is generally suitable for modeling and forecasting analysis.
The overall score combines integrity and forecastability to provide a general data quality judgment. It can be used to quickly decide whether data is ready for the next step and to identify datasets that should be prioritized for governance.
| Score | Meaning | Suggestion |
|---|---|---|
| 80–100 | Data quality is good and suitable for modeling and forecasting analysis | Use with priority |
| 60–79 | Data quality basically meets requirements, and better results can be achieved after governance | Improve based on the weak dimension |
| 0–59 | The data does not meet high-quality modeling requirements, and forecasting results may be unstable | Complete governance before modeling |
If the overall score is low, the data usually has clear weaknesses in integrity or forecastability. Review the detailed results together with the four basic quality metrics to locate specific issues, then perform data governance.
Evaluation Records
The evaluation records list shows historical results for each dataset, including the four basic quality metrics, integrity, forecastability, overall score, creation time, and duration. A metric is displayed as ”-” when no result is returned.
The following actions are available on the right side of each record:
- Click View to open the evaluation details page.
- Click Delete to remove an evaluation record that is no longer needed.
View Evaluation Details
Click View in the evaluation records to open the details page. The page includes evaluation progress, score trends, and object-level details.
- Evaluation progress: Shows the task status and overall progress. When evaluation is complete, the status is Completed and progress is 100%.
- Score trends: Shows the current integrity and forecastability scores and how they changed across evaluations of the same dataset.
- Object details: Uses a Table → Device → Measurement hierarchy to show object names, series counts, total points, integrity, forecastability, and overall scores.
Click the expand icon to view child objects and their scores. You can export the results as CSV for further analysis or switch the table to full-screen view.
Quick Reference
| Metric | Preferred direction | Primary purpose |
|---|---|---|
| Timestamp accuracy | Higher is better | Check timestamp formats, ordering, and duplication |
| Missing value rate | Lower is better | Measure gaps and missing data |
| Stable series ratio | Higher generally means more stable | Assess overall series stability |
| Anomaly rate | Lower is better | Identify outliers and abnormal fluctuations |
| Integrity | Higher scores are better | Determine whether data and key fields are complete |
| Forecastability | Higher scores are better | Determine whether the series contains forecastable patterns |
| Overall score | Higher scores are better | Determine whether the data is suitable for modeling and forecasting analysis |