This study explores the use of graph-based techniques to analyze complex household data from the Ministry of Social Development and Human Security (MSDHS) of Thailand. By employing heterogeneous graphs, the project addresses the challenges of capturing multi-relational, high-dimensional data that traditional methods often struggle to handle. The system utilizes metapath2vec, a node embedding algorithm, to generate low-dimensional vector representations that preserve both structural and semantic relationships within the network. These embeddings facilitate machine learning tasks such as clustering and classification. The dataset, a simulated version of MSDHS’s real-world data, consists of 174,232 nodes in different types of nodes, representing household attributes and member attributes such as income level, family problems, and health conditions. The generated embeddings achieved an F1-score of 0.9856 from a multi-class classification task, demonstrating high quality in predicting household fragility. The embeddings were subsequently used for the household analysis applying the K-means clustering algorithm. The results demonstrate the system’s ability to uncover socio-economic fragility patterns with greater accuracy and interpretability.