విషయ సూచిక · 29 / 35
Read Radar Unusual Activity
నవీకరణ 22 ఆగస్టు 2026 · 3 నిమిషాల చదువు · 14 విషయ సూచిక
సరళమైన అర్థం
Radar Unusual Activity is a direction-neutral ranking of listings whose frozen feature mix was unusual relative to other listings on the same exchange and completed session. Unusual means statistically different within that cross-section; it does not mean bullish, bearish, popular or influential.
Why it is useful #
The board can surface observations worth investigating while preserving coverage, drivers, price basis, source, evidence date and generated time. It is intentionally separate from Radar profile ranking so an anomaly cannot silently promote or suppress a rule-based result.
Where it appears and the workflow #
Open Radar, then Unusual Activity, choose NSE or BSE and review the latest available run. Check the run date, feature coverage and leading drivers before opening a listing. Compare the ranked observation with its source-backed chart and filings; do not act on the rank alone.
Calculation or source #
The separate ML worker reads immutable feature snapshots for one exchange/session/schema/method, median-imputes missing cross-sectional features while retaining each row's original coverage, applies robust scaling using median and dispersion, and persists one natural run. The public route is DB-only and does not train or score on request.
Prerequisites, inputs, period and unit #
Prerequisites are Machine Learning and Unusual Activity both enabled, the ML worker running, a matured NSE or BSE feature snapshot, and a sufficiently usable same-exchange cross-section. Scores and percentiles are dimensionless; driver values keep their disclosed units. Evidence date and generated time are different timestamps.
Worked example #
Illustrative only: a listing ranks at the 96th percentile in an NSE run with 19 of 22 features available. That means its measured feature mix was more unusual than most rows in that frozen run; it does not say the next return will be positive or that the three missing features were zero.
What high and low mean #
A higher percentile means more unusual within that exact run, not better, worse or more certain. A large positive or negative standardized driver shows which side of the cross-sectional centre contributed to the anomaly; the sign is not a trade direction.
Positive, negative and zero #
A standardized feature can be positive, negative or genuinely zero relative to the run centre. A zero anomaly contribution can also arise from a flat feature. Missing original inputs remain disclosed through coverage and must not be interpreted as observed zero values.
Limitations and common mistakes #
Cross-sectional ranks change when the eligible universe, session, schema or peer distribution changes. Median imputation can preserve a row in the run but does not create source evidence. Common mistakes include reading the board as volume alone, comparing percentiles across different runs, or calling an anomaly a prediction.
Market-specific differences #
Unusual Activity currently accepts NSE and BSE only and fits them separately. No result from one venue is relabelled as the other. Availability on the general Radar product does not imply this worker-dependent board is enabled, and other exchanges remain unsupported rather than approximated.
Suggested next steps #
First verify the exchange, evidence date, generated time and coverage. Open the top row, list its disclosed drivers and independently inspect their source evidence. Repeat for a middle-ranked row so the percentile is read as relative context, then record why no direction follows from the comparison.
Educational use only #
This guide explains descriptive research evidence. It is not investment advice, a price prediction, a recommendation, a claim of predictive accuracy, or an instruction to buy, sell, rebalance or place an order.