In the context of venture capital (VC), violation risk prediction of high-tech enterprises has become a non-negligible component of investment decision-making. Due to the high degree of uncertainty in technological innovation, market volatility and policy environment of high-tech enterprises, traditional violation risk prediction methods are often difficult to cope with multiple risk sources and data privacy across institutions involved. Transfer learning, as an emerging machine learning method that can migrate knowledge from the source domain to the target domain, provides the novel ideas to solve these multiple risk sources. Therefore, based on the transfer learning framework, this study proposes a novel violation risk prediction model for high-tech enterprises by using multi-source information: auditors, regulators, firms, social media, and VC institutions. The proposed cross-industry transfer learning model (CTLM) first extracts valuable knowledge from multi-source information and then transfers it to the target domain-violation risk prediction-to enhance the accuracy and robustness of the predictions. We construct a comprehensive dataset from Wind and CVSource databases spanning 2000-2023, encompassing seven dimensions including VC exit histories (N = 1000 + samples per source domain), shareholding structures, financial metrics, legal proceedings, audit opinions, regulatory inquiries, and market sentiment. The data undergo rigorous preprocessing through standardization, categorical encoding, missing value imputation, and SMOTE oversampling to address class imbalance (1:4 positive-negative ratio). In the empirical results, validated through fivefold stratified cross-validation and 80/20 train-test split across seven distinct knowledge transfer scenarios, both the simulation experiments and real-world experiments, all validation show that the transfer learning model, CTLM, can effectively improve the performance of violation risk prediction and provide more accurate support for VC decisions.