Choose the time field and value series
Drop a CSV, TSV, JSON, JSONL, XLSX, or XLS file into the page. After it loads, choose Timestamp column and one or more fields under Value columns (series). The detector sorts usable points by timestamp, then compares each point with earlier points in that series.
- Drop a CSV, TSV, JSON, JSONL, or spreadsheet file here, or click browse files.
- Choose the time field in Timestamp column.
- Select one or more numeric fields in Value columns (series).
- Set Duplicate timestamps, Flag gaps larger than (x typical interval), Method, and Threshold. Adjust Rolling window (prior points) or the seasonal settings.
- Review the summary, charts, flagged rows, then click Export CSV or Export JSON.
What each detection method compares
Choose Method for the baseline used by each series. Rolling z-score calculates the mean and standard deviation of the prior Rolling window (prior points). Rolling IQR calculates prior quartiles, uses the median as Expected, and flags values outside the Threshold multiplied by the interquartile range.
Seasonal residual compares a point with the same phase in earlier cycles. Set Season period (points per cycle) and Prior cycles to compare. The method needs at least two periods, then uses the median and median absolute deviation of those prior values.
timestamp,value
Method: Rolling z-score
Rolling window (prior points): 20
Threshold: 3
Timestamp | Series | Observed | Expected | Score | Threshold | ReasonHow timestamps, duplicates, and gaps are handled
Timestamp column supplies the ordering field. In Timestamp display, choose Parse as-is (literal ISO instant), Interpret & display as UTC, or Interpret & display in browser local time. Invalid timestamps, blank values, missing-value tokens, and non-numeric values are excluded before scoring.
Duplicate timestamps can use Keep first occurrence, Keep last occurrence, Average duplicate values, or Flag every duplicate as an anomaly. The selected rule runs after sorting. Flag gaps larger than (x typical interval) compares each interval with the series' median interval. The summary counts qualifying gaps, and a flagged point immediately after one can include a gap note in Reason.
Points without enough prior history are skipped. A constant baseline window has zero spread, so its point cannot receive a score. These cases appear under Skipped (not scored).
What the results show and export
Each selected series gets a line chart with flagged points marked. The table lists Timestamp, Series, Observed, Expected, Score, Threshold, and Reason. The page displays the first 500 flagged points; Export CSV and Export JSON include the complete flagged set and Method.
What a flagged point can include
Every flagged row keeps the original observed value beside the expected value, score, threshold, and reason. Duplicate rows flagged by the selected rule have no baseline score. Their reason identifies the duplicate timestamp handling rule.
A flagged point after a qualifying gap can carry a note that its baseline may use limited recent history. The chart keeps the series in timestamp order, so the point can be checked against the interval before it.
The detector analyzes existing observations. It does not forecast future values, fill gaps, repair values, interpolate, or modify the source file.
When another dataset analysis tool fits
Use this page within Dataset Analyze Tools when time order and prior history define an unusual observation. Choose the Outlier Detector when you need high or low values in numeric columns without a timestamp-based baseline.
Choose the Dataset Profiler for column types, row and column counts, null and blank rates, duplicates, date ranges, and data-quality warnings. Choose the Correlation Matrix Generator when the task is pairwise Pearson, Spearman, or Kendall correlation between numeric columns.
Frequently asked questions
How do I detect anomalies in time-series data?
Drop in the data, choose Timestamp column and Value columns (series), then set Method, Threshold, and the relevant window or cycle settings. The tool scores usable observations and reports flagged points with Expected, Score, and Reason.
Which rolling anomaly detection method should I choose?
Rolling z-score compares with the prior mean and standard deviation. Rolling IQR compares with prior quartiles and spread. Choose Seasonal residual for same-phase values in prior cycles; it needs at least two periods.
How does the time-series detector handle duplicate timestamps?
It applies the rule selected in Duplicate timestamps: Keep first occurrence, Keep last occurrence, Average duplicate values, or Flag every duplicate as an anomaly. Invalid timestamps, blank values, missing-value tokens, and non-numeric values are excluded before scoring.
Does the detector change or upload my source data?
It parses and scores the selected file in your browser and leaves the source data unchanged. It reports anomalies, charts the selected series, and creates CSV or JSON downloads for flagged results.