The dataset has several kinds of quality problem: inconsistent formats, mixed units stored in text, duplicate IDs, blanks, and some implausible values. The 23 columns are mostly VARCHAR, so numeric fields could not be used as-is. **1. Inconsistent formats and mixed units (the biggest issue)** - Most numeric measures are stored as text with units or symbols attached: - Age appears as `28 yrs` but also as a bare `97`. - Monthly_AI_Cost appears as `$25`. - Monthly_Income appears as `$1,884`, `0` and `$0`. - AI_Usage_Hours_Per_Day appears as `1.7 hrs`. - Time_Saved_Hours_Per_Week appears as `5 hrs`. - Income has both `0` and `$0` as zero values. All of these need stripping and casting before any averaging. - Gender has 12 spellings for what is roughly 3–4 categories: - Male, male, MALE and M - Female, female, FEMALE and F - Other, other and Non-binary - blank - Education_Level has 17 variants for about 6 levels. Examples are `Bachelor's`, `Bachelors`, `bachelor's degree` and `BA/BSc`; `Master's`, `Masters`, `master's degree` and `MA/MSc`; `PhD` and `Ph.D.`; `High School`, `high school` and `HS`; and `Undergraduate` and `undergrad`. `Undergraduate` and `Bachelor's` may also overlap in meaning. - The tool listing was cut off at 50 rows, so I could not see every AI_Tool variant. The column range shows a lowercase `gemini`, which suggests `Google Gemini` appears under more than one spelling. - Different measures use different scales: - Accuracy_Rating runs 2–5. - Satisfaction_Score runs 3–10. - Productivity_Score runs 1.6–10.0 and is fractional. They are not directly comparable without rescaling. **2. Duplicates** - There are 300 rows but only 288 distinct User_IDs, so about 12 rows repeat an ID. The grain is supposed to be one row per user, so these need de-duplicating or investigating. **3. Missing values** - Blank counts by column: | Column | Blank or null | |---|---| | Age | 23 | | Productivity_Score | 16 | | Accuracy_Rating | 15 | | Profession | 15 | | Education_Level | 13 | | AI_Purpose | 13 | | Monthly_AI_Cost | 12 | | AI_Tool | 11 | | Gender | 10 | | Country | 10 | | Monthly_Income | 10 | | Time_Saved_Hours_Per_Week | 10 | | Work_or_Study_Hours_Per_Day | 9 | | AI_Usage_Hours_Per_Day | 7 | | Satisfaction_Score | 7 | | Would_Recommend | 5 | - Missing values are stored as empty strings in the text columns, so a plain null check would miss them. **4. Outliers and impossible or implausible values** - **Age:** it ranges from 16 to 97, and `97` appears twice in a column that otherwise uses `NN yrs`. A bare 97 may be a typo or placeholder for a different value. - **Income:** the maximum is 250,000 against a median of about 521.5. That is a very skewed distribution, and the maximum is probably a currency or unit mismatch or an entry error. - **Zero income:** 34 rows have income of 0. This is plausible for students, but it should be confirmed that 0 does not mean "missing". - **AI cost:** the maximum is 100, while the most common non-zero values are $20–$50. Most cost values are `$0` or blank. - **Daily AI usage:** the maximum is 23 hours per day. - **Usage versus work or study hours:** in 24 rows, daily AI usage exceeds work or study hours per day. The maximum work or study figure is 20, and 2 rows exceed 16 hours per day. - **Time saved versus usage:** the maximum time saved is 19.4 hrs per week, and 1 row reports more time saved per week than total weekly AI usage. **Recommendation:** before analysis, parse the numeric text columns, standardise Gender, Education_Level and AI_Tool to canonical values, de-duplicate on User_ID, treat empty strings as null, and flag or cap the extreme income, usage and age values.