The economy-water system involves complex interactions between water resources and economic activities; fully understanding these complexities and balancing economic development with water resources consumption remains a significant challenge. This study develops a factorial multi-region input-output CNN-LSTM (FMCL) model through integrating deep learning (convolutional neural network and long short-term memory, CNN-LSTM) and multivariate statistical analysis (factorial design, FD) into a multi-region input-output (MRIO) framework. The FMCL model not only unveils the socioeconomic drivers underlying water consumption patterns, but also quantitatively assess complicated interactions within the economy-water system under multiple scenarios. The FMCL model is then applied to a typical water-scarce region in the Yellow River Basin (three provinces of Inner Mongolia, Shaanxi and Ningxia, abbreviated as Inner-Shaan-Ning region). The main findings are: (i) direct water consumption intensity, industrial structure (agriculture, manufacturing, construction, and accommodation & catering), and per capita household consumption constitute the key socioeconomic drivers of water consumption; (ii) among different shared socioeconomic pathways (SSPs), Inner Mongolia, Shaanxi, and Ningxia would achieve relatively high levels of economy-water sustainability under SSP2, which represents a pathway of moderate socioeconomic development trends; (iii) the interaction between accommodation & catering and per capita household consumption would be significant; compared to the baseline SSP2, interactive policies that focus on key factors would achieve higher economy-water sustainability; (iv) from 2026 to 2050, the optimal interactive policy in Inner-Shaan-Ning region could reduce annual water consumption per unit of GDP by 1.22%, 3.82% and 3.34%, respectively. These findings cannot only reveal the key drivers and their multi-dimensional interactions within the economy-water system, but also facilitate the transformation of water use pattern into sustainable development.