Skip to contents

RHISEA

“R Package for Mixed-Stock Analysis Extending HISEA with Flexible and Robust Classification Frameworks”

RHISEA modernizes and extends HISEA with flexible, robust classification frameworks to estimate source contributions and assign individuals to their stock of origin, with rigorous uncertainty quantification.

Key features:

  • Drop-in compatibility with classic HISEA (FORTRAN) workflows, plus modern extensions
  • Multiple classification approaches (parametric and nonparametric), or use your own model and apply HISEA estimators
  • Estimation of stock proportions and individual assignment with uncertainty via bootstrap, cross-validation, and simulations
  • Built-in diagnostics for transparent interpretation Designed for fisheries biologists, ecologists, managers, and analysts seeking robust, reproducible inferences to support decision-making

Table of contents

Installation

RHISEA can be installed from GitHub:

library(devtools)
install_github("SostheneA/RHISEA", dependencies = TRUE)

Citation

To cite RHISEA in publications use:

citation("RHISEA")

From HISEA FORTRAN to RHISEA: Modern Mixed-Stock Analysis in R

Mixed-stock analysis is a fundamental tool for fisheries science, allowing estimation of the contributions of source populations to mixture samples (e.g., fisheries catches). The original HISEA program, implemented in FORTRAN, pioneered this approach with a robust statistical framework and introduced several estimators (Θ₁ to Θ₅) for stock proportion inference.

However, the classical HISEA had important limitations:

  • Only a single built-in classifier (LDA/discriminant analysis), requiring the variables to be approximately multivariate normal with constant covariance matrices.
  • Limited flexibility for integrating modern statistical learning approaches.
  • Rigid input/output based on text files (.std, .mix, *.ctl), and no option to save intermediate or diagnostic results for downstream analysis.

The Modern Workflow: RHISEA in R

RHISEA builds on the foundation of HISEA and brings mixed-stock analysis into the present:

The Modern Workflow: RHISEA in R

RHISEA builds on the foundation of HISEA and brings mixed-stock analysis into the present:

Key Innovations

1- Flexible Classifier Choice

With RHISEA, you are not limited to LDA. Built-in support now exists for a wide range of modern supervised learning classifiers:

  • Linear/Quadratic Discriminant Analysis (LDA, QDA, MASS)

  • Mixture models (MCLUST)

  • Decision Trees (CTREE, CART)

  • Linear/Multinomial Regression (MLR)

  • k-Nearest Neighbors (KNN, kknn)

  • Support Vector Machines (SVM)

  • Random Forests (RF, randomForest)

  • Extreme Gradient Boosting (XGB)

  • Artificial Neural Networks (ANN, nnet)

  • Naive Bayes (NB, e1071)

    …and you can even supply your own trained model and prediction function.

2- Modern Data Handling

You can work directly with R data frames—no need for fixed-format text files. However, full compatibility is retained: you may still use HISEA-formatted files for reproducibility or validation against legacy outputs.

3- Expanded Outputs & Reproducibility

  • RHISEA logs not only the final report but can save all intermediate objects (classification probabilities, confusion matrices, etc.) as .rda files for downstream diagnostics and reproducibility.
  • Results are summarized with confidence intervals via bootstrap, and automated visualizations are provided.
  • All run parameters, models, and derived files can be archived for full transparency and reproducibility.

4- Intuitive, Modular Workflow

Each step of the process is transparent: from baseline/mixture data ingestion → classifier choice and fitting → estimation with all classical HISEA estimators (Θ₁: Raw, Θ₂: Cook & Lord, Θ₃: Cook Constrained, Θ₄: EM MillarC, Θ₅: EM ML) → summary and report. You control every aspect or automate it all.

Step-by-Step Overview (see diagram)

  1. Start: Specify your inputs—choose between R data frames or HISEA .std/.mix control files.
  2. Classifier Selection: Easily select or define your choice of classification model for stock assignment.
  3. Model Training & Validation: Fit the classifier to the baseline data, estimate the classification matrix (Φ), and validate model accuracy/confusion.
  4. Application to Mixture: Predict class probabilities for the mixture, obtain individual assignments and posterior probabilities.
  5. Estimation: Apply the suite of HISEA estimators to convert classification outputs to stock proportion estimates (Θ₁–Θ₅).
  6. Summary & Reporting: Save a comprehensive summary report, with options to log all intermediate files (.rda).
  7. End: Ensure results are reproducible, documented, and ready for interpretation or downstream analysis.
HISEA workflow diagram
HISEA workflow diagram

Conclusion: The RHISEA Advantage

With RHISEA, you retain all the statistical rigor of original HISEA - including compatibility and reproducibility with legacy workflows - while gaining the flexibility, transparency, and analytic power of modern R. Advanced users can plug in custom classifiers, automate cross-validation, and leverage the full R ecosystem for further statistical or graphical analysis. All intermediate steps are traceable and archivable for robust, audit-friendly science.

“Reference: Millar, R.B. (1990),”A Versatile Computer Program for Mixed Stock Fishery Composition Estimation”, Can. Tech. Rep. Fish. Aquat. Sci. 1753: iii + 29 p.”