Customer segmentation is crucial for retailers as it involves grouping customers into distinct clusters based on shared characteristics or behaviours. By identifying homogeneous groups of customers who exhibit similar traits or respond similarly to marketing efforts, companies can tailor their strategies and messages more precisely, increasing effectiveness and engagement. This study aimed to develop and analyse machine learning models for effective customer segmentation using historical transactional data from a retail chain, as well as to model the transition probabilities between identified segments. The dataset included features such as customer ID, purchase date, and gross transaction value. It was pre-processed and transformed into RFM (Recency-Frequency-Monetary) metrics to provide meaningful input for clustering algorithms. Several machine learning models were tested, including Agglomerative Clustering, BIRCH (Balanced Iterative Reducing and Clustering using Hierarchies), K-Means, and Random Trees Embedding. The best segmentation results were obtained using the K-Means algorithm, which accurately identified 9 customer clusters. Each cluster was defined as a segment based on observed behavioural patterns. In addition to profiling segment characteristics, the study proposed a graph-based model to estimate the probabilities of customer transitions between segments. This adds practical value by enabling businesses to anticipate changes in customer behaviour and adapt strategies accordingly. Overall, the research provides actionable insights into customer behaviour and offers data-driven recommendations for targeted marketing and customer relationship management.