Scholarlink

<p>The global video game industry has become a data-intensive and highly competitive domain, where understanding the determinants of sales performance is vital for strategic decision-making. Yet, existing studies often remain fragmented, lacking temporal modeling, interpretability, and integrated exploratory analysis. This study introduces an interpretable hybrid machine learning framework for forecasting global video game sales by combining ensemble learning, temporal forecasting, and explainable artificial intelligence (XAI). Using a curated multi-regional dataset of approximately 700 globally released titles enriched with critic evaluations, user ratings, platform adoption levels, online search intensity, and seasonal release indicators, the framework merges advanced gradient-boosting algorithms, light gradient boosting machine (LightGBM), categorical boosting (CatBoost), and extreme gradient boosting (XGBoost), with a Prophet–XGBoost hybrid model to capture both nonlinear dependencies and temporal dynamics. The hybrid model achieved the highest predictive accuracy (root mean square error ≈ 0.148; R<sup>2</sup> ≈ 0.867), improving performance by 9–14% over standalone ensemble and statistical baselines. Post-hoc explainability analyses showed that content quality, hardware reach, and digital visibility collectively account for more than 70% of the predictive variance, underscoring their dominant influence on market outcomes. Principal component (PC) and cluster analyses revealed stable cross-regional archetypes, mainstream Western, genre-specialized Japanese, and heterogeneous emerging markets, validated through silhouette and variance-ratio metrics. Robustness tests confirmed model consistency under feature removal, temporal segmentation, and parameter perturbations. Overall, the proposed framework provides a replicable, end-to-end analytical tool that unites predictive precision with interpretability, supporting evidence-based planning for release timing, platform strategy, and marketing investment.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Data-driven and explainable forecasting of global video game sales: a hybrid machine learning approach

  • 期刊论文

  • 发表日期：

    2026年5月13日

查看全文
  • Omid Abdolazimi,
  • Sina Salamat Mostaghim,
  • Mustafa Rezaei,
  • Junfeng Ma

摘要

The global video game industry has become a data-intensive and highly competitive domain, where understanding the determinants of sales performance is vital for strategic decision-making. Yet, existing studies often remain fragmented, lacking temporal modeling, interpretability, and integrated exploratory analysis. This study introduces an interpretable hybrid machine learning framework for forecasting global video game sales by combining ensemble learning, temporal forecasting, and explainable artificial intelligence (XAI). Using a curated multi-regional dataset of approximately 700 globally released titles enriched with critic evaluations, user ratings, platform adoption levels, online search intensity, and seasonal release indicators, the framework merges advanced gradient-boosting algorithms, light gradient boosting machine (LightGBM), categorical boosting (CatBoost), and extreme gradient boosting (XGBoost), with a Prophet–XGBoost hybrid model to capture both nonlinear dependencies and temporal dynamics. The hybrid model achieved the highest predictive accuracy (root mean square error ≈ 0.148; R2 ≈ 0.867), improving performance by 9–14% over standalone ensemble and statistical baselines. Post-hoc explainability analyses showed that content quality, hardware reach, and digital visibility collectively account for more than 70% of the predictive variance, underscoring their dominant influence on market outcomes. Principal component (PC) and cluster analyses revealed stable cross-regional archetypes, mainstream Western, genre-specialized Japanese, and heterogeneous emerging markets, validated through silhouette and variance-ratio metrics. Robustness tests confirmed model consistency under feature removal, temporal segmentation, and parameter perturbations. Overall, the proposed framework provides a replicable, end-to-end analytical tool that unites predictive precision with interpretability, supporting evidence-based planning for release timing, platform strategy, and marketing investment.