Prediction of reinforced concrete column capacities by machine learning

The concept of machine learning is becoming more popular day by day and is the subject of new fields of study. In this context, it has also found a place for itself in the field of Structural Engineering. Whether the column parameters predicted by machine learning can be an alternative to traditional methods has been examined within the scope of the thesis study. The database of the study consists of 520 column test specimens with rectangular or square cross-sections. Prediction models created using regression, k nearest neighbor, support vector machines, decision trees, random forest, XGBoost, AdaBoost, artificial neural networks, Naive Bayes, gradient boosting and CatBoost machine learning algorithms; The maximum shear strength of rectangular or square columns, the drift ratio corresponding to the point where the maximum shear strength decreases by %20 after the point where it starts to decrease, and the test result are has been used to predict the damage type. The performances of the estimation models were examined using the root mean square error, coefficient of determination, adjusted coefficient of determination, k-fold cross validation and confusion matrix. It has been observed that the maximum shear strength prediction with an accuracy of 93.7%, the translation ratio corresponding to the point where the maximum shear strength starts to decrease by 20% is estimated with 70.1%and the damage type is estimated with 97% accuracy. In addition, an interface that performs estimation using the algorithm with the highest estimation performance has been developed so that users can easily make predictions