Package {PrismR}

Detect Potential Data Leakage

Description

Identifies potential sources of data leakage prior to predictive modeling. The assessment includes identifier columns, duplicate columns, high-cardinality features, predictors identical to the target, and predictors showing an almost perfect correlation with the target.

Usage

detect_leakage(data, target = NULL)

Arguments

Value

A named list containing:

identifier_columns

Names of columns identified as potential identifiers.

duplicate_columns

Names of duplicated columns.

high_cardinality_columns

Names of high-cardinality columns.

target_leakage

Names of predictors identical to the target variable.

correlation_leakage

Numeric predictors with near-perfect correlation to the target.

leakage_score

Overall leakage score ranging from 0 to 100.

variables

A data frame containing variable-level leakage diagnostics, including identifier, duplicate, high-cardinality, target leakage, correlation leakage, and leakage score.

Examples

# Supervised leakage detection
res <- detect_leakage(airquality, target = "Ozone")
res$leakage_score
res$correlation_leakage

# Unsupervised check for IDs and duplicates
res_unsupervised <- detect_leakage(iris)
res_unsupervised$identifier_columns

Feature Stability and Distribution Drift Assessment

Description

Evaluates feature distribution drift between reference and current datasets (or across sequential time partitions of a single dataset) using the Population Stability Index (PSI) and scale-normalized Wasserstein distance across both numeric and categorical variables.

Usage

feature_stability(
  data,
  current = NULL,
  split_ratio = 0.5,
  partition_col = NULL,
  verbose = FALSE
)

Arguments

Details

Feature drift is evaluated using standard industry thresholds:

• Stable (PSI < 0.10): Insignificant distributional shift.

• Moderate Drift (0.10 <= PSI < 0.25): Noticeable shift; monitoring recommended.

• Unstable / Severe Drift (PSI >= 0.25): Severe distributional divergence; retraining or feature review required.

For categorical variables and discrete bins, Bayesian Laplace smoothing is applied to prevent division-by-zero errors when frequencies are zero in one partition.

If only one dataset is supplied, the function performs a sequential partition-based stability assessment (comparing the first partition to the second partition of observations). For production monitoring, users should provide reference and current datasets separately via data and current.

Value

A list containing:

stability_score

Overall stability score from 0 to 100.

verdict

Categorical stability verdict: "Stable", "Moderate Drift", or "Unstable".

variables

A data frame of feature-level diagnostics containing variable, type, psi, wasserstein, stability_score, and status.

unstable_features

Character vector of features flagged with severe drift (PSI >= 0.25).

moderate_features

Character vector of features flagged with moderate drift (0.10 <= PSI < 0.25).

stable_features

Character vector of features exhibiting stable distributions (PSI < 0.10).

method

Character string describing the assessment method used ("two_sample_comparison" or "sequential_partition").

Examples

# Single-dataset sequential drift evaluation
res <- feature_stability(airquality)
res$stability_score
res$verdict

# Comparing reference baseline vs current dataset
ref <- data.frame(val = rnorm(100, 0, 1))
cur <- data.frame(val = rnorm(100, 2, 1))
res2 <- feature_stability(ref, current = cur)
res2$verdict

Plot a Prism Report

Description

Visualises the results of a Prism Analysis using one of several diagnostic plots.

Usage

## S3 method for class 'PrismReport'
plot(
  x,
  y = NULL,
  type = c("radial", "circular", "bubble", "radar"),
  feature = NULL,
  ...
)

Arguments

Details

Available visualisations include:

• radial: Radial/gauge chart summarising the overall health of each PrismR module and dataset readiness.

• circular: Feature Diagnostic Rose (coxcomb) displaying feature Completeness (petal length), Redundancy & Leakage Status (petal color), and Recommended Transformation (outer rim cap) around an open donut center.

• bubble: Statistical feature map displaying Data Completeness (100% - Missing%) versus Distribution Skewness, with leakage vector colors and transformation shapes.

• radar: Spider/radar chart displaying the raw statistical diagnostic profile of a selected variable (Completeness, Symmetry, Tail Normalcy, Uniqueness, Leakage Safety, and Feature Stability).

Value

A ggplot object.

See Also

prism

Examples

report <- prism(airquality, target = "Ozone")
plot(report, type = "radial")
plot(report, type = "circular")
plot(report, type = "bubble")
plot(report, type = "radar", feature = "Solar.R")

Print a PrismReport

Description

Print a PrismReport

Usage

## S3 method for class 'PrismReport'
print(x, ...)

Arguments

Value

Invisibly returns the input PrismReport object.

Examples

report <- prism(airquality, target = "Ozone")
print(report)

Perform Complete Prism Statistical Validation

Description

Evaluates dataset readiness for predictive modeling across four foundational statistical pillars: data quality, data leakage, distribution transformations, and feature stability. Produces an overall Model Readiness score (0-100) and a gatekeeping verdict.

Usage

prism(data, target = NULL, current = NULL)

Arguments

Details

The four evaluation dimensions include:

• Data Quality: Evaluates missingness, duplicate rows, and constant features.

• Data Leakage: Identifies identifier keys, duplicate features, high-cardinality columns, and target correlation leakers.

• Transformations: Evaluates skewness and excess kurtosis to recommend variance-stabilizing transforms (Box-Cox, Yeo-Johnson, Log).

• Feature Stability: Assesses covariate distribution drift via Population Stability Index (PSI) and Wasserstein distance.

Value

A PrismReport S3 object containing:

quality

List of data quality metrics returned by quality_score.

leakage

List of data leakage diagnostics returned by detect_leakage.

transformation

List of transformation recommendations returned by recommend_transform.

stability

List of stability metrics returned by feature_stability.

readiness

Overall composite Model Readiness score (0-100).

verdict

Gatekeeping verdict: "Model Ready", "Proceed with Caution", or "Action Required".

Examples

report <- prism(airquality, target = "Ozone")
print(report)
summary(report)

Data Quality Assessment

Description

Computes basic data quality metrics for a dataset, including missing values, duplicate rows, constant columns, and an overall quality score.

Usage

quality_score(data)

Arguments

Value

A named list containing:

missing_percent

Percentage of missing values.

duplicate_rows

Number of duplicated rows.

constant_columns

Names of constant columns.

n_constant_columns

Number of constant columns.

quality_score

Overall quality score (0-100).

variables

A data frame containing variable-level quality metrics, including missing percentage, constant feature indicator, and quality score.

Examples

res <- quality_score(airquality)
res$quality_score
res$missing_percent

Recommend Feature Transformations

Description

Analyses numeric variables in a data frame and recommends suitable transformations based on skewness and excess kurtosis. The function does not modify the data; it only provides transformation recommendations to assist with preprocessing before statistical modelling or machine learning.

Usage

recommend_transform(data)

Arguments

Details

Recommendations are based on:

• Distribution skewness.

• Excess kurtosis (tail heaviness).

• Whether the variable contains non-positive values.

Possible recommendations include:

• None

• Log

• Box-Cox

• Yeo-Johnson

Value

A list containing:

recommendations

A data frame of features requiring transformation (excluding "None").

n_numeric

Number of numeric variables analysed.

n_recommended

Number of variables for which a transformation was recommended.

variables

A complete data frame of all analysed numeric features and their metrics.

Examples

recommend_transform(iris)

Summarize a PrismReport

Description

Summarize a PrismReport

Usage

## S3 method for class 'PrismReport'
summary(object, ...)

Arguments

Value

Invisibly returns the input PrismReport object.

Examples

report <- prism(airquality, target = "Ozone")
summary(report)