Comparative Evaluation of Transition Network Analysis (TNA) and Frequency Transition Network Analysis (FTNA) Models

Supplementary material to the paper "Frequency Transition Network Analysis (FTNA)" accepted at ISLS 2025. This report is a systematic evaluation that was conducted using 10,000 synthetic datasets designed to mimic real-world complexities. These datasets varied in the number of states, the number of rows (representing independent sequences), the length of each sequence, and the presence of missing values. The number of sequence spells per dataset ranged from 300 to 2,000 to assess the impact of sample size on edge weight estimation. Sequence lengths were set between 15 and 100 states per row to account for both short and long temporal patterns. Furthermore, a random number of missing values (between 0 and 20 per sequence) was introduced to evaluate the robustness of estimations under incomplete data conditions, which reflects the variability often found in real-world educational datasets with varying session sizes. The objective of this report is to analyze the results of this evaluation assessed with four primary categories of metrics: weight deviation, correlation, similarity, and dissimilarity measures (Saqr, Tikka, and López-Pernas 2024). The analysis was performed using the compare function of the tna package with minmax scaling to ensure both networks are compared on the same scale.