Predicting the Financial Performance of Startups Using Reinforcement Gradient Algorithms and Model Explainability Analysis Based on Shapley Value Indicators

Abstract

The objective of this study was to predict the financial performance of startup firms using a reinforcement gradient learning algorithm while explaining model predictions through Shapley value–based explainable artificial intelligence indicators. This applied quantitative study employed a predictive analytics design integrating machine learning and explainable artificial intelligence. The statistical population consisted of technology-oriented startups operating in Tehran, from which 162 active startups were selected through purposive sampling based on operational continuity, financial transparency, and data availability. Financial and operational data covering the period 2019–2023 were collected from audited reports, accelerator databases, and innovation ecosystem records. The dependent variable was a composite financial performance index derived from revenue growth, profitability, cash flow stability, and investment efficiency measures. Independent variables included digital engagement, innovation investment, funding diversity, organizational growth indicators, and human capital characteristics. Data preprocessing involved normalization, missing value imputation, and outlier adjustment. A reinforcement gradient algorithm was developed for prediction and optimized using cross-validation procedures. Model interpretability was examined through Shapley value analysis to quantify the contribution of each predictor to financial performance outcomes. Results indicated that the reinforcement gradient model achieved high predictive accuracy (R² = 0.89) and significantly outperformed traditional regression and ensemble learning approaches. Shapley value analysis revealed that revenue growth rate, digital engagement, research and development investment, and funding diversification were the strongest contributors to predicted financial performance. The model demonstrated stable generalization across validation samples, confirming the effectiveness of reinforcement learning in capturing nonlinear relationships among entrepreneurial, financial, and technological variables. Explainability results further showed heterogeneous performance pathways among startups, indicating that successful financial outcomes emerged from integrated combinations of innovation capability, digital maturity, and strategic resource management rather than single-factor effects. The findings demonstrate that combining reinforcement gradient algorithms with explainable artificial intelligence provides a powerful and transparent framework for forecasting startup financial performance, offering valuable insights for investors, entrepreneurs, and policymakers seeking evidence-based decision support in dynamic entrepreneurial ecosystems.