Logarithmic Incremental Parameter Tuning of Support Vector Machines for Human Activity Recognition

Objectives: In machine learning based human activity monitoring, the algorithm needs to produce a prediction model with a high accuracy. Support vector machine is one of the leading machine learning algorithms. Methods/Statistical Analysis: We propose an optimization approach of support vector machines that optimizes its regularization parameter to further improve its prediction accuracy in a human activity recognition application. In order to implement an efficient support vector machines predictive model of a particular dataset that would generalize well and have a good prediction performance, a suitable regularization parameter has to be applied in the regularization part of the equation. Findings: In order to empirically evaluate the effectiveness of our proposed approach, we show the results of our implementation and discuss the results of our proposed approach explained in the previous section on support vector machines models. From our experiments, we can see that we got fabulous results when the regularization parameter is 1000. For the accuracy on train/test dataset pair, we got a sufficiently high percentage for regularization parameter values of 10, 100 and 1000. And, the best cross validation accuracy is 98.8575, which is corresponding to a regularization parameter value of 1000. Additionally, we can also notice that the relation between the classification accuracy and the cross validation accuracy is proportional, and that is obvious in the accuracies responding to the regularization parameter of 0.0001, because both accuracies are significantly low. Improvements/Applications: Our idea was to replace the parameter value with a vector of parameter values and compare their results. It shows more improved and promising performance improvement but if we can apply parallel programming.