With growing interest in Multi-Criteria Decision Analysis (MCDA) as a powerful tool for decision-making in various fields, there is a need for reliable and accessible resources for MCDA researchers and practitioners. To address this need, an open data repository for MCDA-related resources has been developed. The goal of this repository is to provide a centralized platform for sharing and accessing MCDA resources, enabling researchers and practitioners to collaborate, replicate studies, and advance the field of MCDA. The repository is open to contributions from the MCDA community, ensuring its continual growth and usefulness.

The dataset contains pairwise comparison matrices (PCMs) that have been manipulated using various attacks, making it an ideal resource for the development of methods and models designed to detect such manipulations. Along with the matrices themselves, the archive also includes a Python script that was used to generate the data, which can be useful for researchers who want to replicate the dataset or explore different methods of analysis. With this dataset, researchers can test and refine their approaches to identifying and mitigating manipulation in pairwise comparison data, ultimately leading to more accurate and reliable results in a variety of fields.

DOI: 10.58032/AGH/R8O0EJ

Download: https://doi.org/10.58032/AGH/R8O0EJ

n our research, we have undertaken a significant task testing the vulnerability of opinion aggregation algorithm in the Montecarlo method. This is a crucial study area, directly impacting decision-making and data analysis. Our paper, authored by K. Kulakowski, J. Szybowski, J. Mazurek, and S. Ernst (2024), titled 'Resilient heuristic aggregation of judgments in the pairwise comparisons method, 'published in Information Sciences 657 pp. 119979, presents our findings. We used a comprehensive set of 80000 pairwise comparison matrices, each corresponding to a pairwise comparison of n alternatives made by a specific decision agent (decision maker). Our research process was intricate, involving the formation of groups of 20 decision-makers who share a similar level of inconsistency and preference profile. These groups, formed by larger sets, are united by a similar preference profile, but the levels of preference consistency within these groups vary. The data we present reflects this complexity, with 40 levels of preference consistency among groups of agents. Ultimately, we drew 100 preference profiles from the studied data, with 34 profiles for five alternatives, 33 for six alternatives, and 33 for seven alternatives. Hence, in the file containing the test data in JSON format, we find a three-dimensional array whose elements are matrices corresponding directly to individual agents' preferences. At the attribute level labeled agent_matrix, we have individual elements—preference matrices. Higher up, at the level labeled agent_group, we have 20-person groups of agents with similar preference profiles and similar inconsistencies. In essence, these groups are being manipulated. At another level higher (labeled agent_profile), we have sets of groups corresponding to a similar decision profile with a given number of alternatives. Finally, we have a list of sets corresponding to different decision profiles at the highest level of the array, which is the input data set (labeled list_of_profiles).

DOI: 10.58032/AGH/KPOHPT

Download: https://doi.org/10.58032/AGH/KPOHPT

A distinct set of the training data was generated for each of the tested size of the PC matrix C. Each set consists of three distinct .json files, which are used to train and test the neural networks designed to detect one of the three described attack algorithms (naive, basic, advanced). The previously mentioned .json files have the same name as the algorithm they are used to test (e.g. basic.json), and are placed in a folder whose name determines the size (n) of the matrix C and the number of independent data samples for each of the tested algorithms (e.g. data_7_7_10000).

DOI: 10.58032/AGH/E8LV5J

Download: https://doi.org/10.58032/AGH/E8LV5J

The set of files contains a matrix of weights, where each vector is a set of weights (ratings) from one expert. In each case, there are exactly as many experts as evaluated alternatives, hence each matrix is a square matrix. The dataset is composed of records containing the following fields: id - stands for number of record, W - forms a n \times n matrix, and ei - is a column of a matrix containing weights comming from i-th expert. There are ten files in the dataset having names in the form: k2_data_100000_matrixSize_XX.json where XX stands for both the number of experts and alternatives.

DOI: 10.58032/AGH/LPCQ56

Download: https://doi.org/10.58032/AGH/LPCQ56

Unpacked data.zip archive contains several .json files with data. Each file contains data used to test two pairwise comparison matrix attack algorithms, following the article https://doi.org/10.48550/arXiv.2211.01809 referred to as "row" and "hadamard". The name of each of the files contains information of the matrix size n and the distance between the promoted and reference alternatives (e.g. 7x7_delta_3, where 7 is the matrix size, and 3 is the delta).

DOI: 10.58032/AGH/NHXCHT

Download: https://doi.org/10.58032/AGH/NHXCHT

The files contain pairwise comparison matrices (PC matrices) created by the anonymized respondents. Respondents had a general knowledge of the pairwise comparison method. They also understood the general regularity according to which the better the results an alternative receives in individual comparisons, the higher weight it will receive in the final prioritization vector. The file 7x7_hones.json contains 84 7 x 7 matrices for a survey in which all respondents were asked to answer the questions posed honestly. The file 7x7_dishonest.json contains 80 7 x 7 matrices for a survey in which all respondents were asked to answer the questions posed insincerely. The file 9x9_mixed_1.json contains 75 expert response records containing matrices of size 9 x 9 for a survey in which all respondents were asked to answer honestly or not honestly. Each record assigned to an expert additionally contains an “honesty” field with values of 0 or 1. Zero indicates an insincere - manipulated - response. 1 indicates a sincere response. The file 9x9_mixed_2_inc.json contains 70 expert response records containing a 9 x 9 matrix for a survey in which all respondents were asked to respond honestly or insincerely. Each of the records assigned to an expert additionally contains an “honesty” field with values of 0 or 1. Zero indicates an insincere - manipulated - response. 1 indicates a sincere answer. The data for one answer was corrupted, creating an incomplete matrix. Therefore, the record assigned to the expert has an additional is_complete tag indicating whether the matrix is complete or not. The missing space in the incomplete matrix was filled with a value of 0.

DOI: 10.58032/AGH/A4HWVZ

Download: https://doi.org/10.58032/AGH/A4HWVZ

Pairs is a proof-of-concept application for iOS for collecting responses in a way that follows the Analytic Hierarchy Process (AHP), see https://en.wikipedia.org/wiki/Analytic_hierarchy_process. Pairs is written in Swift, using iOS SDK, and uses SwiftUI for its user interface. The application language of GUI is Polish. Pairs requires near-zero configuration; it uses a WebDAV (https://en.wikipedia.org/wiki/WebDAV) server specified in the configuration to retrieve the poll definition and sends the responses back to the server. For a more detailed description, read the readme file.

DOI: 10.58032/AGH/BT8OQM

Download: https://doi.org/10.58032/AGH/BT8OQM