Comparison of k-nearest Neighbor & Artificial Neural Network prediction in the mechanical properties of aluminum alloys

Abstract

Discovery of new materials is increased after the introduction of high accuracy machine learning techniques in the field of material science. Traditional way of discovering new composites requires lot of experimentation and laboratory measurements which consumes more energy and time. Property prediction using latest machine learning technology which saves lot of energy and time by avoiding experimentation and laboratory measurements. As a result, many researchers utilized the data analysis techniques to predict the properties of different material with high accuracy. So, in this research, the methodology is developed to predict the mechanical properties such as yield strength, hardness and tensile strength using ANN (Artificial Neural Network) and KNN (K-Nearest Neighbour algorithm). In Tensile strength and Hardness prediction, KNN has the higher R square value and Lower RMSE value compared to ANN which leads to more accurate predictions. But in the case of Yield strength prediction, ANN has the high R square value and Low RMSE value. The testing data size is varied from10 to 50 percent and RMSE values are calculated to prove the above results and also to find optimum testing data size. The suited algorithm is selected based on the R square and RMSE values and test size also varied to find the optimum test size of data to achieve more accuracy in prediction.