We investigate people’s mobility patterns in tourist cities through WiFi data, focusing on higher-order homophily in hypergraphs. We first develop a method for detecting Points-of-Interest (POIs) in a tourist city from WiFi data, and examine the higher-order interactions among the detected POIs using a POI hypergraph, which is constructed by treating the group of POIs visited by an individual in one day as a hyperedge. Aiming to understand how the predictability of people’s flow, based on locally restricted observations, influences the formation of the POI hypergraph, we introduce two entropy-based predictability measures for the POIs. For each measure, we classify the POIs into three classes indicating high, medium and low predictability, and propose a method for extracting important POIs for each of these class labels in terms of homophily. Using WiFi data from Kyoto, we reveal several interesting homophily properties of the POI hypergraph, and show that the proposed method can identify interesting POIs in terms of homophily that are difficult to extract by traditional graph-based methods dealing with pairwise interactions.