Implementing Machine Learning in Laboratory Synthesis by Hybrid of SVR Model and Optimization Algorithms

Abstract Numerous studies have been performed to modify ferrites to achieve the desired magnetic properties for the intended applications. Many variables with interactions between themselves are involved in modifying these properties. This study aims to provide an accurate and effective method for predicting the magnetic properties of M‐type ferrites under the influence of involved variables. For this purpose, ferrites are synthesized by doping 17 different ions and the results of the analysis of samples form the experimental data. The support vector regression (SVR) model is selected for prediction and in order to optimize hyper parameters, genetic, particle swarm optimization, and sequential minimal optimization techniques are used. Based on the comparison between the outcomes of these algorithms, the model with the best performance is used to predict coercivity and residual magnetism in M‐type magnetic ferrites. Furthermore, with the cross validation technique, the accuracy and robustness of the model designed for new samples are evaluated. The results show that the selected SVR model, with an average absolute relative error of 2% and 0.7%, is able to predict coercivity and residual magnetism, respectively. Consequently, the prediction method presented has the capability to accelerate and develop the modification of magnetic properties for different applications.