This PhD study is based on considering the sequential nature of match activity occurrences to understand the (physical) demand of games on players as well as uncover players’ behavioural movement patterns during competitive matches.
It involves quantifying players’ external load (i.e., completed match activities) using movement patterns that provide information on how players accumulated external load in contrast to existing physical and technical-tactical performance indicators.
The quantification of external load helps with team and player improvement, training specificity and even talent identification and recruitment.
Elite players of rugby football league were considered as the participants of this PhD study because rugby league is a physically intense team sport where activities happen quickly and within a stipulated timeframe.
Existing player movement profiling frameworks were investigated which revealed the use of a sequential pattern mining algorithm to extract movement patterns.
However, the algorithm for extracting patterns in the existing player movement profiling frameworks is found to have limitations such as the identification of few movement patterns, providing only the longest common patterns within a cluster of movement sequences while it discards other interesting patterns.
Furthermore, a review of sequential pattern mining algorithms revealed none of the existing algorithms is suitable for player movement profiling.
The state-of-the-art algorithm for extracting closed contiguous patterns (i.e., CCSpan) does not scale well on large data as well as data with long rows of sequences.
Also, the CCSpan algorithm is without a parameter to define a maximal length of extract patterns which is useful as a sliding window.
This PhD study's contributions to the body of knowledge include the development and optimisation of a novel pattern mining algorithm for extracting user-defined length of frequent closed contiguous patterns, called LCCspm (l-length closed contiguous sequential pattern mining), among others.
An intrinsic evaluation of the LCCspm algorithm was considered in terms of speed and memory consumption performance measures.
Results revealed it is four, seven or ten times faster than the state-of-the-art algorithm when tested on natural data in three different use cases.
An extrinsic evaluation of the LCCspm algorithm reveals and validates that its movement patterns are best to profile players into playing positions when compared to other distinct types of obtainable movement patterns.
Furthermore, this PhD study applied LCCspm and other artificial intelligence algorithms to discover signature movement patterns of elite rugby league players per playing position, identify (key) movement patterns as predictor variables for classifying players into playing positions, and later establish a set of movement performance indicator useful for assessing players’ performance variability across playing positions.
In the broader scope of computing, this PhD study contributes LCCspm algorithm as an advancement of pattern mining algorithms.
The application of LCCspm can be extended to other sports domains and challenges, allowing for more accurate analysis and insights into player behaviour and team dynamics.
Additionally, LCCspm can be applied in any field where the consecutiveness of items matters during the analysis of patterns.