I calibrated a model to my dataset and I want to use parameters of this model as features. This model have parameter alpha that is periodic so for signal with period 'S' I will have alpha_1, alpha_2,.... and alpha_S. Because S is different for each of learning data, the number of features become different but for classification same number of features is necessary.
I have used mean and standard deviation of alpha_1,... , alpha_S as feature but the result wasn't good. Does anybody have an idea?
Feature selection becomes prominent, especially in the data sets with many variables and features. It will eliminate unimportant variables and improve the accuracy as well as the performance of classification. Random Forest has emerged as a quite useful algorithm that can handle the feature selection issue even with a higher number of variables. In this paper, we use three popular datasets with a higher number of variables (Bank Marketing, Car Evaluation Database, Human Activity Recognition Using Smartphones) to conduct the experiment. There are four main reasons why feature selection is essential. First, to simplify the model by reducing the number of parameters, next to decrease the training time, to reduce overfilling by enhancing generalization, and to avoid the curse of dimensionality.
Dear Maryam Hamidi
The chi-squared statistic and the mutual information statistic are the two most often used feature selection methods for categorical input data where the target variable is likewise categorical (e.g., classification predictive modeling).
Recursive Feature Elimination (RFE) accepts as input a Machine Learning model instance and the final number of features to employ. It then iteratively decreases the amount of features to employ by ranking them based on the accuracy of the Machine Learning model as metrics.
The only way PCA is a legitimate approach of feature selection is if the most relevant variables happen to be the ones with the highest variance. After completing PCA, you will get uncorrelated variables that are a linear combination of the previous variables.
The goal of feature selection approaches is to minimize the number of input variables to those that are thought to be most beneficial to a model in order to predict the target variable. The goal of feature selection is to remove non-informative or redundant predictors from the model.
Kind Regards
Qamar Ul Islam
Please note that you can construct vectors of features.
Then to know how close one vector to the other you can do projection.
This can tell you
the correlation between
the two vectors.
Thanks for all your comments. You have mentioned feature selection and feature reduction methods. Let I explain more about my question. Assume there are just 3 records as training data and their periods are 10, 17 and 15 respectively. The model has one periodic parameter (alpha) so after calibrating I obtain 10, 17 and 15 alphas respectively for records 1, 2 and 3. If I want to use model's parameters as feature, I will have 10 features for record 1, 17 features for record 2 and 15 features for record 3. Can I apply feature reduction methods?
PCA can extract m features from n features (m
- Badges
- Science topic
- Similar topics
- Engineering
- Materials Engineering
- Powders