This study pairs unsupervised learning methods of clustering and classification with discrete-time Markov Chain models for airport pavement performance prediction, with the goal of allowing the clustering algorithm to take the place of typical pre-analysis grouping of pavement sections. The influences of cluster method, number of clusters, input years, interval time span, and training data on prediction results are analysed. Results indicate that clustering using the deterioration profiles of Pavement Condition Index (PCI) before applying the Markov Chain is sufficient to reduce the error by half in predicted pavement conditions for most cases. The initial pavement degradation is a reliable source for clustering and future condition prediction using the appropriately matched Markov Chain. This approach minimises the data required to predict pavement condition and proves the early deterioration rate of a pavement section can be used to predict its overall lifespan.