Hybrid Feature-Based Critical Success Factors in Cloud Enterprise Resource Planning Through Artificial Neural Networks and Random Forest

Cloud enterprise resource planning (CERP) systems are widely adopted to enhance operational efficiency.However, in a global context, ERP implementation failure rates remain notably high, ranging from 67% to 90%.If left unaddressed, they will hinder sustainable economic growth and industrial transformation.To address this issue, this study adopts a structured approach by identifying critical success factors (CSFs) based on key performance indicators (KPIs) using the Delphi method and Dempster-Shafer combination method.The resulting dataset integrates multi-stage CSFs and their associated KPI performance and weight values, forming a hybrid feature set that captures interrelated implementation factors.The effectiveness of each implementation stage is assessed through user feedback scores categorized as satisfactory (>4), below satisfactory (3.0-3.9), and failure (<3.0).To evaluate the predictive capability of this hybrid dataset, both artificial neural network (ANN) and random forest (RF) models were applied separately.Each model was trained and tested independently to identify which algorithm achieves higher prediction accuracy for implementation outcomes.The findings indicate that RF significantly outperforms ANN, with an accuracy of 0.849 compared to 0.765.Additionally, confusion matrix, ROC, and AUC analyses further confirm RF's superior predictive capability.Through this research, the identification of CSFs through qualitative analysis or statistical modelling, combined with the integration of machine learning techniques, ultimately improving assessment classification for CERP implementation.