Performance evaluation of hybrid nanofluid-filled cylindrical heat pipe by machine learning algorithms

Abstract

The current study attempts to predict the outlet temperature of a hybrid nanofluid heat pipe using three machine learning models, namely Extra Tree Regression (ETR), CatBoost Re-gression (CBR), and Light Gradient Boosting Machine Regression (LGBMR), in the Python environment. Based on 7000 experimental data (various heat input, inclination angle, flow rate, and fluid ratio), different training (95%–5%) and testing (5%–95%) split sizes, a closer prediction was attained at 85:15. The three attempted machine learning models are capable of predicting the outlet temperature, as evidenced by the less than 5% deviation from the experi-mental results. Of the three attempted machine learning models, the ETR model outperforms the other two with a higher accuracy (98%). Further, the sensitivity analysis indicates the ab-sence of data overfitting in the attempted models.