Forecasting Organizational Innovation Performance Through Long Short-Term Memory (LSTM) Networks and Strategic Capability Indicators

Abstract

Objective: The objective of this study was to forecast organizational innovation performance using Long Short-Term Memory (LSTM) networks based on strategic capability indicators and to evaluate the predictive contribution of key organizational capabilities to future innovation outcomes.

Methods and Materials: This quantitative longitudinal predictive study was conducted among 312 organizations operating across multiple industries in Canada, including manufacturing, information technology, healthcare, financial services, telecommunications, and professional services. Data were collected from 1,248 senior managers and executives using standardized instruments measuring strategic flexibility, technological capability, organizational learning capability, knowledge management capability, market sensing capability, innovation capability, absorptive capacity, resource integration capability, and leadership capability. In addition, longitudinal organizational performance records covering five consecutive years were obtained from organizational databases and annual reports. Data preprocessing procedures included normalization, outlier treatment, and temporal sequence generation. The dataset was divided into training (70%), validation (15%), and testing (15%) subsets. Organizational innovation performance was forecasted using a Long Short-Term Memory neural network. Comparative analyses were conducted using Linear Regression, Support Vector Regression, Random Forest, XGBoost, and Multilayer Perceptron models. Model performance was evaluated using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²). SHapley Additive exPlanations (SHAP) analysis was employed to determine predictor importance.

Findings: The results revealed significant positive correlations between all strategic capability indicators and organizational innovation performance (p < .01). Technological capability demonstrated the strongest association with innovation performance (r = .77), followed by organizational learning capability (r = .74) and knowledge management capability (r = .71). Comparative model evaluation indicated that the LSTM network outperformed all alternative forecasting approaches, achieving the lowest prediction errors (RMSE = 0.167, MAE = 0.129, MAPE = 3.74%) and the highest explanatory power (R² = 0.947). SHAP analysis identified technological capability, organizational learning capability, and knowledge management capability as the most influential predictors of future innovation performance. Furthermore, the LSTM model maintained strong predictive accuracy across all industrial sectors, with R² values ranging from 0.932 to 0.958, demonstrating robust cross-sector generalizability.

Conclusion: The findings demonstrate that organizational innovation performance can be forecasted with high accuracy through LSTM-based deep learning models utilizing strategic capability indicators. Technological capability, organizational learning, and knowledge management emerged as the most critical determinants of future innovation outcomes. The superior performance of the LSTM model highlights the importance of capturing temporal and nonlinear relationships within organizational data and suggests that artificial intelligence-driven forecasting systems can provide valuable support for strategic planning, innovation management, and long-term organizational decision-making.