Dynamic Pattern Analysis: Method Overview and Trajectory Assessment of Object Development

Procedia Computer Science. 2026. Vol. 287. P. 193–200.

Myachin A. L.

We extend the static pattern analysis method to the temporal dimension by introducing a six-type trajectory taxonomy that classifies objects according to the frequency and structure of pattern switches over an observation window of T > 8 periods. For each object, a reference pattern is designated as the most frequently occupied group over the observation history, and a Bayesian membership scoring formula is applied at each time point using interval-encoded binary features under the Bernoulli naive Bayes assumption. The score tracks, over time, how well the object’s feature vector continues to match its reference group. Key properties of the formula are examined: normalization is guaranteed by construction; log-posterior monotonicity depends on whether the interval covers the modal region of each feature within the pattern; the expected average score decomposes analytically across trajectory types; and the computational cost is O(N . T . n . K). A synthetic experiment with T = 8, three Gaussian patterns, and N = 300 objects confirms that the proposed scoring formula separates trajectory types substantially (η2 = 0.38), that the observed mean ordering of scores is consistent with Property 7, and that the same formula without reference-pattern anchoring scores near chance (η2 = 0.007).

Keywords: dynamic pattern analysis, trajectory classification, reference pattern