Comparative Evaluation and Comprehensive Analysis of Machine Learning Models for Regression Problems

Abstract Artificial intelligence and machine learning applications are of significant importance almost in every field human life to solve problems or support experts. However, the determination model achieve a superior result for particular problem within wide real-life application areas is still challenging task researchers. The success could be affected by several factors such as dataset characteristics, training strategy responses. Therefore, comprehensive analysis required determine ability efficiency considered strategies. This study implemented ten benchmark models on seventeen varied datasets. Experiments performed using four different strategies 60:40, 70:30, 80:20 hold-out five-fold cross-validation techniques. We used three evaluation metrics evaluate experimental results: mean squared error, absolute coefficient (R2 score). analyzed, each model's advantages, disadvantages, data dependencies indicated. As excess number experiments, deep Long-Short Term Memory (LSTM) neural network outperformed other models, namely, decision tree, linear regression, vector regression with radial basis function kernels, random forest, gradient boosting, extreme shallow network, network. It has also been shown that tremendous impact results experiments should studies where mining selection not performed.