Time-Series Prediction of Sales, Yield, and Fitness Performance

Time-series prediction plays a critical role in commerce, agriculture, and health monitoring. This review examines published research on forecasting models across three domains: sales volume and price, crop yield, and fitness performance. In sales forecasting, statistical approaches such as ARIMA remain effective for stable seasonal demand, while ensemble methods including Random Forest and XGBoost demonstrate improved accuracy on retail datasets, and attention-based recurrent neural networks reduce error on multi-regional produce data. In crop yield prediction, hybrid machine learning frameworks incorporating climatic and agronomic variables highlight the superiority of ensemble regressors, especially Random Forest and Bagging, in modeling nonlinear input-output relationships. In fitness performance analysis, supervised learning methods applied to wearable sensor data confirm the robustness of ensemble algorithms, with Random Forest achieving lower mean absolute error compared to decision tree and support vector machine baselines. A cross-domain synthesis reveals that ensemble tree models consistently deliver competitive results in heterogeneous data settings, while recurrent neural networks are particularly suited for sequential dependency in sales data. The evidence underscores that model effectiveness depends on domain-specific covariates, data resolution, and temporal structures. This survey consolidates quantitative results from multiple studies to provide a comparative reference for future methodological choices in time-series research.