# Column classification: `AI_Usage_and_Impact_on_Students_and_Professionals` (23 columns) This is a partial answer. It rests on the profiler's metadata (type, inferred role, distinct counts, ranges). I did not inspect raw values, so some calls below are judgment, not confirmed. ## Identifier | Column | Evidence | |---|---| | **User_ID** | Tagged `identifier` and `near_key`. About 287 distinct values, 0 nulls, range U0001..U0288. It is *near*-unique rather than a clean key, so there may be a duplicate or gap (288 IDs implied vs ~287 distinct). | `Tasks_Performed` has about 288 distinct values and is tagged `attribute`/`descriptive`. It is free text, not an ID. ## Categorical dimensions | Column | Distinct (approx.) | Note | |---|---|---| | Gender | ~12 | Many variants for what should be a few values, so labels likely need cleaning (range ends in "other"). | | User_Type | ~2 | Professional / Student. | | Education_Level | ~17 | Likely inconsistent spellings (e.g., "undergrad"). | | AI_Tool | ~16 | e.g., "gemini". | | AI_Purpose | ~7 | e.g., "Writing". | | Would_Recommend | ~4 | Likely Yes/No plus variants. | | Country | ~20 | Tagged `spatial`, so it is a geographic dimension. | | Profession | ~110 | Tagged `spatial`, but that looks like a misclassification. It is really a high-cardinality categorical, with values like "Theoretical Physics Student". | | Tasks_Performed | ~288 | Free-text descriptor, not useful as a grouping dimension. | ## Numeric metrics **Already stored as numbers:** | Column | Type | Nulls | Range | |---|---|---|---| | Productivity_Score | DOUBLE | 16 | 1.6–10.0 | | Satisfaction_Score | BIGINT | 7 | 3–10 | | Accuracy_Rating | BIGINT | 15 | 2–5 (only ~4 distinct values, so it behaves like an ordinal rating) | **Numeric in meaning but stored as VARCHAR (0 nulls):** - Age (~51 distinct) - Monthly_Income (~232 distinct, max shown 9938) - Monthly_AI_Cost (~15 distinct, max shown 50) - AI_Usage_Hours_Per_Day (~81 distinct). The range shows a value like "9.9 hrs", so unit text is mixed in. - Work_or_Study_Hours_Per_Day (~99 distinct) - Time_Saved_Hours_Per_Week (~118 distinct) The minimums of these text columns show as blank, which suggests empty strings or non-numeric entries. They need cleaning and casting before aggregation. ## Dates/times - **_ingestion_timestamp** (TIMESTAMP) is the only temporal column. It has 1 distinct value (2026-10-02 09:58:27), so it is load metadata, not an analytical time dimension. - The source data has no event-date column, so time-trend analysis isn't possible. ## Pipeline/metadata columns `_batch_id`, `_source_file` and `_source_system` each have 1 distinct value. They are ingestion lineage and carry no analytical value. ## Uncertain - I haven't verified whether the VARCHAR numeric columns cast cleanly. - The profiler's role tags are imperfect. Examples: Profession as "spatial", and Age labelled "measure" even though it is more of a demographic attribute or bin. - I did not check what causes the near-key status of User_ID.