Comparative analysis dust accumulation impact on PV performance using artificial neural network and machine learning algorithms

• Comparative analysis of ANN and SVR models for predicting PV power output under dust. • Experimental PV power prediction with aerosol impact on dust accumulation studied. • ANN and SVR models evaluated on RMSE, MAPE, and R-value using PV experimental data. • SVR outperforms ANN with RMSE (0.24), MAPE (1.54), and accuracy (98.3 %). This paper presents a comparative study on artificial neural networks (ANN) and machine learning (ML)-based modelling approaches on experimental data to predict the power output of photovoltaic (PV) systems with the aerosol impact on different types of dust samples such as chalks powder, fly ash, rice husks, sand and bricks powder accumulation on the PV module surface. An experimental study deliberates about the maximum power output through the 60 W PV module during the artificial irradiation levels (625 W / m 2 , 675 W / m 2 , 725 W / m 2 , 825 W / m 2 , and 875 W / m 2 ) with different dust samples and weights. Both the ANN and support vector regression (SVR) models are trained verified on sample data generated from PV module under controlled laboratory environment. The PV module power output is predicted in terms of root mean square error (RMSE), mean absolute percentage error (MAPE), and R-value to compare the ANN and SVR models' performance. In this context, ANN-based performance metrics in terms of RMSE (1.41), MAPE (11.011), R-Value (0.983), and accuracy (97.02 %) are on the lower side compared to the SVR model as RMSE (0.24), MAPE (1.544), R-Value (0.995), and accuracy (98.3 %). Future utility-scale PV power plants may utilize ANN and SVR-based models for real-time monitoring, predictive maintenance, manual cleaning, and on-site diagnostics. These models complement the experimental setup and provide a scalable, futuristic predictive framework for autonomous, data-driven next-generation solar energy system operation.