archive api data sample

Overview

Source: /home/coolhand/html/datavis/data_trove/joshua-project/archive/api_data_sample.json

Saturn profiled 50 rows across 107 columns. The stats below are deterministic and machine-readable; the prose is a language-model interpretation of those stats (opt-in, added after the fact, never sees raw rows).

ROL3 categorical feature

ROL3 holds 3-letter codes (likely ISO 639-3 language tags such as 'eng', 'arz', 'ary', 'apc', 'afb') across 50 complete rows with 10 distinct values. Distribution is skewed toward Levantine/Gulf Arabic variants: 'apc' covers 36% and 'afb' another 13/50, while six codes appear only once or twice. Entropy ratio of 0.75 indicates moderate concentration rather than uniform spread.

Treatment: Group rare codes into an 'other' bucket, then one-hot encode.

PhotoCredits categorical metadata

Attribution string for the image accompanying each row, naming photographer and source platform. Just 5 distinct credits cover all 50 rows, with 'Hashim Abdullah - Pixabay' alone accounting for 56% and the top three Pixabay/Flickr contributors covering 48 of 50 entries. Two credits ('Link Up Africa', 'Claudiovidri - Shutterstock') appear only once, suggesting a long tail of incidental sources.

Treatment: Retain as provenance metadata; drop from modelling features.

PrimaryReligion categorical feature

Categorical column capturing the dominant religion of each record, with only two observed values across 50 rows. The distribution is severely imbalanced: Islam accounts for 49 of 50 entries (top_rate 0.98) and Christianity for just 1, yielding an entropy ratio of 0.14. With effectively no variance, this column carries almost no discriminative signal.

Treatment: Drop or collapse to a binary 'is_Islam' flag; near-constant for modelling.

Ctry categorical feature

Country field with 41 distinct values across 50 rows and no nulls. The distribution is essentially flat — entropy ratio is 0.986 and the most common value, United Arab Emirates, appears just twice (4%). Nine countries tie at 2 occurrences each, the rest are singletons, hence the long_tail alert.

Treatment: Group rare countries into regions or an 'Other' bucket before encoding.

RegionName categorical feature

RegionName is a categorical geographic grouping with 11 distinct regions across 50 rows and no nulls. The distribution is uneven: 'Africa, North and Middle East' alone accounts for 32% (16/50), and the three African regions together dominate the column. Entropy ratio of 0.86 indicates spread is fairly even given the cardinality, but several regions ('Australia and Pacific', 'America, Latin') appear only once.

Treatment: one-hot or target-encode; consider grouping single-row regions to avoid sparse levels.

BibleYear categorical metadata

BibleYear appears to be a metadata field capturing the publication year or year-range of a Bible edition, with values like "1890-2024", "1382-2020", and "2021". The column is almost entirely empty: 92% null with only 4 of 50 rows populated and 3 distinct values. Format is inconsistent, mixing single years with hyphenated ranges, which blocks numeric parsing.

Treatment: Drop or quarantine; null rate of 0.92 and mixed year/range formats make it unusable without manual curation.

RLG3PC numeric other

RLG3PC is a numeric column that is entirely constant: all 50 rows hold the value 6.0, with zero variance and only 1 unique value. There is no information for a model to learn from, and no nulls or outliers to caveat.

Treatment: Drop; constant column carries no signal.

Population numeric feature

Population counts across 50 rows, ranging from 200 to 2,221,000 with a median of just 46,500 versus a mean of 264,074. The distribution is severely right-skewed (skew 2.52, kurtosis 5.78) and 12% of rows (6 values) flag as outliers, indicating a few very large populations dominate an otherwise small-town dataset. No nulls or zeros, and 49 of 50 values are unique.

Treatment: log-transform before any modelling to tame the right skew and outliers.

Resources unknown other

The column is named "Resources" and contains 50 non-null entries, but saturn skipped profiling it so its kind is unknown and no descriptive stats (uniqueness, distribution, type) are available. Without further signals, its content and structure cannot be characterized from this evidence.

Treatment: Re-run profiling with type coercion or inspect raw values manually before deciding on use.

LeastReachedPGAC categorical feature

Binary Y/N flag indicating whether some 'least reached PGAC' condition was met, with no nulls across 50 rows. The split is nearly balanced (28 N, 22 Y; top_rate 0.56) and entropy_ratio of 0.99 confirms maximal informativeness for a binary feature.

Treatment: Encode as 0/1 boolean for modelling.

NumberLanguagesSpoken numeric feature

Counts the number of languages spoken, but with only 2 unique values across 50 rows it is effectively a binary indicator of monolingual vs bilingual. The mean of 1.04 and median of 1.0 show 48 of 50 records sit at 1, with just 2 outliers at 2.0 driving the extreme skew (4.69) and kurtosis (20.04).

Treatment: Recode as a binary multilingual flag; the raw count adds no information.

GSEC categorical feature

GSEC is a binary categorical field with exactly two values, "1" and an empty string, split perfectly 25/25 across the 50 rows. The maximum entropy (1.0) confirms a balanced flag, but the empty string rather than "0" or null suggests the absent state is encoded as a blank string rather than a true missing value.

Treatment: Recode empty string to 0 (or NaN if it means missing) and treat as a binary indicator.

AudioRecordings categorical metadata

AudioRecordings is a categorical flag that takes the single value 'Y' across all 50 rows, with zero nulls and entropy of 0. Because cardinality is 1 and top_rate is 1.0, the column carries no information and cannot discriminate between records.

Treatment: Drop; constant column with no variance.

PercentAdherents categorical feature

PercentAdherents holds numeric percentages stored as strings, with 29 distinct values across 50 rows and no nulls. The mode is "0.000" at 12% of rows, and entropy ratio 0.942 indicates a very flat distribution with a long tail of small-frequency values. The mix of fractional (0.200, 0.500) and whole-number (5.000, 6.000) entries suggests the values are raw percentages rather than proportions.

Treatment: Cast strings to float and treat as a continuous numeric feature.

ROP1 categorical metadata

ROP1 is a categorical column holding a single constant value 'A001' across all 50 rows, with zero nulls and entropy of 0.0. It carries no information for modelling or segmentation since cardinality is 1 and top_rate is 1.0.

Treatment: Drop; constant column with no variance.

JPScalePGAC categorical feature

A binary categorical field with only the values "1" and "2", split 28/22 across 50 rows. The near-maximal entropy ratio (0.99) indicates an almost balanced two-class distribution with no nulls. The column name suggests a Japanese PGA scale code, likely an ordinal seismic-intensity or rating bucket.

Treatment: Cast to categorical (or 0/1 indicator) before modelling.

Latitude numeric feature

Numeric column holding geographic latitude in degrees, with all 50 values unique and no nulls. The range spans -33.87 to 59.11, consistent with worldwide coordinates, and the distribution is mildly left-skewed (-0.46) with a mean of 22.28 sitting below the median of 24.12. Only one outlier (2%) is flagged, suggesting one row sits far from the otherwise broad spread (IQR 26.18).

Treatment: Pair with longitude as a geospatial coordinate; consider binning or distance features rather than using as a raw scalar.

PeopNameInCountry categorical label

Categorical label naming the people group in-country, with only 7 distinct values across 50 rows and no nulls. The distribution is heavily concentrated on Arab variants: 'Arab' alone covers 54% of rows, and the top three Arab-prefixed labels account for 46 of 50 entries, leaving a long tail of singletons like 'Tuareg, Air' and 'Amri'. Entropy ratio of 0.65 confirms the imbalance flagged by the long_tail alert.

Treatment: Collapse rare singletons into an 'Other' bucket before any group-wise analysis.

Window1040 categorical feature

Window1040 is a binary Y/N flag, almost perfectly balanced with 26 'Y' and 24 'N' across 50 rows. Entropy ratio of 0.999 confirms a near-maximum-uncertainty split, and there are no nulls. The name suggests a windowed indicator (possibly a 1040-period rolling event flag), but the evidence does not confirm its semantics.

Treatment: Encode as a 0/1 boolean for modelling.

PeopleGroupMapURL categorical metadata

URL pointing to a people-group map image hosted on joshuaproject.net, one per row. 48% of the 50 rows are empty strings, so the field is missing more often than populated, and a single map (m00007.png) accounts for 9 of the 26 non-blank entries. With 17 unique values across 50 rows and a long_tail alert, most distinct URLs appear only once.

Treatment: Treat empty strings as missing and drop from modelling; keep only as a display link.

CountryURL categorical foreign_key

URLs to country pages on joshuaproject.net, with the two-letter country code as the path suffix. With 41 unique values across 50 rows and entropy ratio 0.986, the column is near-unique; the most frequent URL (UAE) appears just twice (top_rate 0.04). The base domain is constant, so the country code is the only informative part.

Treatment: Extract the trailing country code and use that as the join/grouping key instead of the raw URL.

PercentEvangelicalPC categorical feature

Numeric-looking field stored as a categorical with only 3 distinct values across 50 rows, and 96% of rows share the single value '0.197'. The other two values ('0.103' and '0.265') each appear exactly once, giving an extreme imbalance and an entropy ratio of just 0.178. This looks like a principal-component or aggregate score that has been collapsed/repeated for nearly every record, leaving almost no signal.

Treatment: Drop or treat as constant — near-zero variance offers no modelling signal.

CountOfProvinces unknown other

CountOfProvinces was skipped by the profiler, so no type, uniqueness, or distribution stats are available beyond a row count of 50 with no nulls. The name suggests an integer tally of provinces per record, but this cannot be confirmed from the evidence. No further signal is present to flag skew, duplicates, or range.

Treatment: Re-run the profiler on this column to recover type and distribution before deciding how to use it.

PercentEvangelicalPGAC categorical feature

Likely a percentage of evangelical adherents (PGAC denomination grouping) stored as strings rather than floats, with only 5 distinct values across 50 rows. The distribution is severely lumpy: '1.892' covers 56% of rows and the top three values ('1.892', '0.233', '0.023') account for 48 of 50 observations, suggesting these are imputed or default category codes rather than true continuous measurements.

Treatment: Cast to float and treat as a low-cardinality categorical or imputed flag rather than a continuous percentage.

NomadicTypeDescription categorical metadata

This appears to be a descriptive label for a nomadic lifestyle classification, but it carries almost no information in this sample. 90% of rows are null, and the 5 non-null rows all hold the single value 'Agro-Pastoralists' (top_rate 1.0, cardinality 1, entropy 0.0). As-is, the column cannot discriminate between records.

Treatment: Drop: 90% null and only one observed value provides no signal.

MapCredits categorical metadata

MapCredits holds attribution strings for the map associated with each row, citing sources like Joshua Project, GMI, ESRI, and Bethany World Prayer Center. Nearly half the rows (24 of 50) carry an empty string rather than a null, and a single credit to 'Bethany World Prayer Center' covers another 14 rows, leaving only 7 distinct values across the column. The dominance of blanks alongside non-null status is the main surprise — missingness is encoded as empty text, not NULL.

Treatment: Normalize empty strings to nulls and treat as provenance metadata; drop from modelling features.

HasJesusFilm categorical feature

Binary Y/N flag indicating whether each record has an associated 'Jesus Film' resource. The column is complete (null_rate 0.0) with only 2 unique values, heavily skewed toward 'Y' at 82% (41 of 50), leaving just 9 'N' cases.

Treatment: Encode as boolean; expect limited discriminatory power given the 82/18 imbalance.

HowReach categorical free_text

HowReach holds free-text outreach suggestions, likely missionary engagement strategies for various people groups. The column is dominated by empty strings (62% top_rate, 31 of 50 rows blank), and every non-blank value is unique, yielding 20 distinct values across 50 rows with no nulls flagged. Entropy ratio of 0.60 plus the long_tail alert confirm this is essentially sparse prose, not a categorical variable.

Treatment: Treat blanks as missing and tokenize/embed the prose entries rather than one-hot encoding.

PCIslam numeric feature

PCIslam appears to be a percentage measure of Islamic affiliation per record, ranging 25.0 to 100.0 with a median of 95.99 and mean 91.12. The distribution is heavily left-skewed (skew -2.65, kurtosis 7.39) with a tight IQR of 6.55 between Q1 92.93 and Q3 99.48, yet 8 outliers (16%) sit far below that cluster. Most observations are near-saturation while a small tail of low-share records pulls the mean down.

Treatment: Consider a reflected log or beta transform before modelling to tame the left skew and downweight the low-share outliers.

NTYear categorical free_text

NTYear appears to be a free-form annotation about a 'NT' year status, mixing a yes/no flag with single years (e.g. 2005, 1932, 2012) and year ranges (e.g. 1879-1989, 1990-2003). The format is inconsistent: 'Yes' dominates at 62% of non-null entries, while 42% of all rows are null and the remaining cells split across 7 heterogeneous values. This is effectively two or three different fields collapsed into one string column.

Treatment: Split into a boolean indicator and parsed year/year-range fields before use.

RLG4 numeric other

RLG4 is a numeric column that is effectively unusable: 86% of its 50 rows are null, and every one of the remaining values equals 20.0 (min, median, max, std all confirm this). With a single distinct value and no variance, it carries no information for modelling.

Treatment: Drop the column; it is constant where present and 86% null.

AffinityBloc categorical metadata

AffinityBloc is a categorical grouping label, but every one of the 50 rows holds the same value, "Arab World". With cardinality 1 and entropy 0, this column carries no information for distinguishing records in this slice.

Treatment: Drop; constant column with zero entropy.

NaturalName categorical label

NaturalName is a low-cardinality categorical label, likely an ethnic or linguistic group identifier, with 7 distinct values across 50 rows and no nulls. The distribution is heavily concentrated: 'Arab' alone covers 54% of rows, and together with 'Gulf-spoken Arab' (11) and 'Omani Arab' (8) accounts for 46 of 50 records, leaving four singleton categories in a long tail. Entropy ratio of 0.65 confirms the imbalance.

Treatment: Group the four singleton categories into an 'Other' bucket before encoding.

PercentChristianPGAC categorical feature

This column appears to be a percentage-based metric (likely 'Percent Christian' from a PGAC indicator) stored as strings, with only 5 distinct values across 50 rows. The distribution is heavily concentrated: '14.741' accounts for 56% of records, followed by '0.935' (12 rows) and '0.066' (8 rows), suggesting these are repeated category-level constants rather than per-row measurements. The presence of just 5 unique values for what looks like a continuous percentage is suspicious and points to either aggregated/joined reference data or a coarse bucketing.

Treatment: Cast to numeric and verify whether values are constants from a join key before using as a feature.

PrimaryLanguageName categorical feature

This is a categorical column naming a primary language, dominated by Arabic dialects across 10 distinct values in 50 rows. Levantine Arabic leads at 18/50 (36%), followed by Gulf (13) and Omani (8); non-Arabic entries are rare (Swahili 2, plus singletons for Tamajeq, English, etc.). Entropy ratio of 0.75 indicates moderate concentration without a single overwhelming class, and there are no nulls.

Treatment: Group rare non-Arabic and minor dialects into an 'Other' bucket before one-hot encoding.

CountOfCountries numeric feature

Numeric count of countries with only 4 unique values across 50 rows, ranging 1 to 28 with median 28 and mean 19.88. The distribution is heavily concentrated at the maximum (median equals max, Q3 equals max), indicating most rows hit the ceiling of 28 while a minority sit much lower. Negative kurtosis (-1.52) and mild left skew (-0.43) confirm a bimodal-like spread rather than a smooth distribution.

Treatment: Treat as a low-cardinality ordinal or bin into categorical buckets rather than as a continuous variable.

PeopleID2 numeric foreign_key

PeopleID2 is stored as numeric but behaves like a categorical key, with only 3 unique values across 50 rows and an IQR of 0 because Q1, median, and Q3 all equal 111. The mean (115.04) is pulled above the median by 2 outliers reaching up to 307, producing extreme skew (6.85) and kurtosis (44.93). No nulls or zeros are present.

Treatment: Treat as a categorical identifier (not a numeric feature) and left-join on it rather than aggregating.

Summary categorical free_text

Free-text ethnographic summaries describing people groups, with 21 unique values across 50 rows and a 60% top rate driven entirely by empty strings (30 of 50). The non-empty entries are long, prose paragraphs about Tuareg, Arab, and other groups, so this is descriptive content rather than a category. Entropy ratio of 0.61 and the long_tail alert confirm most non-empty values appear only once.

Treatment: Treat empty strings as missing and tokenize/embed the prose before any modelling.

Obstacles categorical free_text

Free-text commentary describing barriers to gospel outreach for various people groups, one paragraph per row. 30 of 50 rows (top_rate 0.6) are empty strings, and the remaining 20 entries are essentially unique, yielding 21 distinct values and entropy_ratio 0.61. Despite the categorical kind, the content is prose passages about Islam, identity, and mission access, not a bounded label set.

Treatment: Treat empty strings as missing and tokenize/embed the prose rather than one-hot encoding.

ROP2 categorical feature

ROP2 is a categorical column with only 3 distinct codes (C0013, C0219, C0019) across 50 rows and no nulls. The distribution is extremely lopsided: C0013 covers 96% of rows while the other two codes appear once each, yielding an entropy ratio of just 0.178. As a near-constant feature it carries almost no signal.

Treatment: Drop or collapse rare levels into 'other'; near-constant column unlikely to aid modelling.

RLG3 numeric feature

RLG3 is a near-constant numeric feature: 49 of 50 rows take the value 6.0 (median, Q1, and Q3 all equal 6.0), with a single outlier at 1.0 producing extreme negative skew (-6.86) and kurtosis (45.02). With only 2 unique values and an IQR of 0, this behaves more like a degenerate flag than a continuous measurement.

Treatment: Drop or binarize as is_not_6; near-zero variance makes it useless for most models.

PercentEvangelical categorical feature

PercentEvangelical appears to be a numeric share of evangelicals stored as strings, with 18 distinct values across 50 rows and no nulls. The distribution is heavily concentrated on small values: '0.000' and '0.500' tie for the mode at 8 occurrences each (16% top_rate), while values up to '2.500' form a long tail. High entropy_ratio (0.88) indicates the mass is spread fairly evenly across the small set of bins despite the long_tail alert.

Treatment: Cast strings to float and treat as a continuous numeric feature.

LeastReached categorical feature

Binary Y/N flag, likely indicating whether some 'least reached' status applies to each record. The class is imbalanced toward Y at 76% (38 of 50), with N covering the remaining 12. No nulls, and entropy ratio of 0.80 reflects the moderate skew.

Treatment: Encode as a 0/1 indicator; consider class-imbalance handling if used as a target.

Continent categorical feature

Categorical continent label with 6 distinct values across 50 rows and no nulls. Asia dominates at 38% (19 rows), followed by Africa (14) and Europe (10), while Australia and South America appear only once each. The skewed distribution and high entropy ratio (0.80) suggest reasonable spread but with a clear Asia/Africa concentration.

Treatment: One-hot encode; consider grouping Australia and South America into an 'Other' bucket given single-row counts.

JPScalePC categorical metadata

JPScalePC is a categorical column that holds the single value "1" across all 50 rows, giving it cardinality 1 and zero entropy. With a top_rate of 1.0 and no nulls, it carries no information and likely represents a constant flag or scale parameter that was never varied in this slice.

Treatment: Drop before modelling; the column is constant.

JPScaleText categorical label

Categorical label describing a Joshua Project reach scale, with 4 distinct levels across 50 rows and no nulls. The distribution is heavily skewed: 'Unreached' covers 76% of rows, while 'Superficially Reached' appears just once, giving an entropy ratio of 0.54. Class imbalance will dominate any model trained on this field.

Treatment: Treat as ordinal categorical and address class imbalance (e.g., stratify or collapse rare levels) before modelling.

SecurityLevel numeric feature

SecurityLevel takes only 3 distinct integer values across 50 rows (min 0, max 2, median 2), so it reads as an ordinal category encoded as a number rather than a continuous measure. The distribution is bimodal-leaning: 42% of rows are zero while the median sits at 2, and the strongly negative kurtosis (-1.90) confirms a flat, multi-peaked shape with no outliers.

Treatment: Treat as an ordinal category (0/1/2) or one-hot encode before modelling rather than using as a continuous numeric.

LRTop100 categorical feature

This is a categorical flag (likely 'is in LR Top 100') that takes the single value 'N' across all 50 rows. With cardinality 1 and entropy 0.0, it carries no information for any downstream model. The 'imbalance' alert here reflects total constancy rather than skew.

Treatment: Drop; constant column with zero entropy.

PrimaryReligionPGAC categorical metadata

This column records the primary religion classification (PGAC), but every one of the 50 rows holds the single value "Islam". With cardinality of 1 and entropy of 0.0, it carries no information for modelling or segmentation.

Treatment: Drop; constant column with zero variance.

PCNonReligious numeric feature

Likely a percentage of non-religious population per row, with 50 records and only 6 distinct values. The distribution is dominated by zeros (zero_rate 0.755) with median and IQR both 0, yet a long right tail pushes the max to 10.0 and flags 12 outliers (24.5%). Skew of 1.89 and one missing value (null_rate 0.02) confirm a sparse, heavily right-skewed feature.

Treatment: Treat as sparse; consider a zero/non-zero binary flag plus log1p transform before modelling.

PhotoCreditURL categorical metadata

This column holds photo credit URLs (Pixabay and Flickr links), but with only 3 unique values across 50 rows it functions as a coarse source tag rather than a per-record citation. The top URL covers 58.3% of non-null rows (28 of 50), suggesting the same stock photo is reused widely. Null rate is 4%.

Treatment: Drop or retain as a low-cardinality source label; not useful as a modelling feature.

PhotoCreativeCommons categorical feature

Binary Y/N flag indicating whether a photo carries a Creative Commons license, with no nulls across 50 rows. The distribution is heavily skewed toward 'N' at 84% (42 of 50), leaving only 8 records flagged 'Y'.

Treatment: Encode as a boolean indicator; note class imbalance if used as a predictor.

PrayForPG categorical free_text

Free-text prayer prompts for unreached people groups, with 60% of the 50 rows being empty strings and the remaining 20 entries each unique multi-sentence paragraphs. The dominance of the blank value (top_rate 0.60) coexists with very high textual diversity among non-blank rows, hence the long_tail alert and entropy_ratio of 0.61. No nulls are recorded, but the empty string is functioning as a missing-value sentinel.

Treatment: Treat empty strings as missing and tokenize/embed the remaining prose if used as a feature.

PeopleGroupPhotoURL categorical metadata

This column holds URLs to people-group profile photos hosted on joshuaproject.net, one per row with no nulls. Despite 50 rows, only 5 distinct images appear, and a single photo (p10375.jpg) covers 56% of records while the top three URLs account for 48 of 50 — suggesting many rows share the same people-group identity rather than being unique entities.

Treatment: Drop for modelling; retain as a display asset or join key to a people-group lookup.

ROG2 categorical feature

ROG2 looks like a regional grouping code with 6 categories (ASI, AFR, EUR, NAR, AUS, LAM), fully populated across all 50 rows. Distribution is moderately concentrated — ASI leads at 38% (19/50) and AFR follows at 14, while AUS and LAM appear only once each. Entropy ratio of 0.80 indicates fairly even spread among the top regions but with thin tails.

Treatment: One-hot encode, but consider collapsing AUS and LAM into an 'Other' bucket given single-row support.

PhotoCCVersionText categorical metadata

This is a categorical license-text field for photo Creative Commons versioning, with only 2 distinct values across 50 rows. 84% of entries are empty strings (42/50), and the remaining 8 carry 'CC BY-NC-SA 2.0' — there are no nulls, just blanks standing in for missing licenses. Entropy ratio of 0.63 reflects this binary, heavily imbalanced split.

Treatment: Recode empty strings to missing and collapse to a binary has_license flag.

Longitude numeric feature

Geographic longitude coordinates spanning -118.3 to 151.2, covering most of the globe's east-west range with all 50 values unique. The distribution is left-skewed (skew -0.77) with a median of 39.4 sitting well above the mean of 27.5, and 7 outliers (14%) flag locations far from the central cluster around Europe/Asia.

Treatment: Pair with Latitude as a 2D geospatial feature; avoid scaling independently.

JPScaleImageURL categorical feature

This column holds URLs to Joshua Project gauge images (gauge-1 through gauge-4), almost certainly a visual encoding of an ordinal progress/status scale. Distribution is heavily skewed: 76% point to gauge-1.png, while gauge-3.png appears only once across 50 rows. With just 4 unique values and no nulls, it's a low-cardinality categorical masquerading as a URL.

Treatment: Extract the gauge digit (1-4) and treat as an ordinal feature rather than keeping the URL string.

OfficialLang categorical feature

This is a categorical column naming an official language, with 21 distinct values across 50 rows and no nulls. The distribution is heavily skewed: 'Arabic, Standard' alone covers 36% of rows, followed by English (8) and French (5), while a long tail of languages like Swahili, Ukrainian, Sinhala and Bulgarian appear only once. Entropy ratio of 0.77 confirms concentration at the top despite the wide vocabulary.

Treatment: Group rare languages into an 'Other' bucket before one-hot encoding.

PhotoPermission categorical feature

Binary Y/N flag indicating whether photo permission has been granted, with no missing values across 50 rows. The distribution is severely imbalanced: 'N' covers 49 of 50 rows (top_rate 0.98) with only a single 'Y', yielding entropy_ratio of 0.14. With one positive case, this column carries almost no discriminative signal in the current sample.

Treatment: Drop or hold aside as a near-constant flag until more 'Y' cases accumulate.

PCHinduism numeric feature

PCHinduism appears to be a per-row count or percentage related to Hinduism, with 95.9% of values being zero and only 3 distinct values across 50 rows. The distribution is extremely sparse and right-skewed (skew 5.96, kurtosis 35.36), with a max of 6.0 standing out as an outlier against a median and IQR of 0. Effectively a near-constant feature with rare nonzero spikes.

Treatment: Binarize to zero/nonzero or drop, given 95.9% zeros and only 3 unique values.

PeopleID3 numeric foreign_key

PeopleID3 is a numeric identifier-like field with only 5 unique values across 50 rows, clustered tightly around 10375-10376 (IQR of 1.0). The distribution is severely left-skewed (skew -5.59, kurtosis 31.2) because at least one value drops to 10208 while the bulk sits near the max of 10378, producing a 20% outlier rate. Despite being typed as numeric, the near-constant range and low cardinality suggest this behaves as a categorical key rather than a measurement.

Treatment: Treat as a categorical identifier; do not use as a continuous numeric feature.

PeopleID1 numeric other

PeopleID1 is flagged as constant: every one of the 50 rows holds the value 10, with zero variance and a single unique value. Despite the 'ID' name, it carries no identifying information and cannot distinguish records. There is no null or outlier activity to interpret.

Treatment: Drop, constant column with no signal.

SpeakNationalLang unknown feature

Saturn skipped this column, so no profiling stats were computed beyond row count (50) and a null rate of 0.0. The name 'SpeakNationalLang' suggests a binary or categorical indicator of whether a respondent speaks the national language, but kind is 'unknown' and n_unique is missing, so the actual value distribution cannot be confirmed from this evidence.

Treatment: Re-profile or manually inspect to determine dtype before use; if binary, encode as 0/1.

PortionsYear categorical feature

PortionsYear is a low-cardinality categorical with 9 unique values across 50 rows and a 16% null rate. Most entries are year ranges (e.g., "1939-2021" at 42.9% and "2009-2024"), but 4 rows contain the string "Yes", indicating a mixed/dirty schema where a boolean answer was recorded in a date-range field. Entropy ratio of 0.70 and a long-tail alert reflect many singleton ranges alongside two dominant values.

Treatment: Clean the type mismatch (separate "Yes" from year ranges), then parse ranges into start/end year numerics before modelling.

PrimaryReligionPC categorical feature

This column records the primary religion of a people-cluster (PC), but every one of the 50 rows holds the value "Islam". With cardinality 1 and entropy 0, it carries no information for distinguishing records in this slice.

Treatment: Drop; constant column with zero entropy.