Comparison of all classifiers trained on the voice data set. The comparison is about how the model performed on unseen data. Various evaluation metrics are compared here. Random Forest seems to give the best results.

SVM models with all three kernels were built to compare which kernel gives the best result. For each kernel, parameter tuning was done to get the best parameters for the model to give optimal output. It was seen that the accuracy on unseen data is almost identical for the models built with linear kernel and polynomial kernel. The accuracy is better with the model with radial kernel. There is a reduction in accuracy as the model was first built on training and fitted on training data for all the 3 scenarios, which is as expected. AUC is more for model with polynomial kernel when compared with that built with a linear kernel. The AUC is highest for the one built with radial kernel. The model with radial kernel is doing better on all metrics.

The data set used for the project is collected from recordings of 30 human subjects captured via smartphones enabled with embedded inertial sensors. Many machine learning courses use this data for teaching purposes. This is a multi-classification problem. The data set has 10,299 rows and 561 columns. For the dataset, 30 people were used to perform 6 different activities. Each of them was wearing a Samsung Galaxy SII on their waist. Using the smartphone's embedded sensors (the accelerometer and the gyroscope), the user's speed and acceleration were measured in 3-axial directions. The sensor's data was used to predict user's activity. The user activities could be one of the six below: Walking, Walking - Upstairs, Walking - Downstairs, Sitting, Standing and Laying. The data can be imported from the following UCI Machine Learning repository http://archive.ics.uci.edu/ml/datasets/Human+Activity+Recognition+Using+Smartphones