Skip to content
This repository has been archived by the owner on Jun 22, 2023. It is now read-only.

In this project, I explored this dataset. This is basically a classification problem. I have tried various classical machine learning models like SVM, K-NN, Decision tree, and logistic regression to get state-of-the-art accuracy. Finally, I have also used deep learning models like multilayer perceptron and 1D Conv-net

License

Notifications You must be signed in to change notification settings

deep-pooja/Human-Activity-Recognition

Repository files navigation

Smart Phone Based Human Activity Recognition

In this project,I explored this dataset. This is basically a classification problem. I have tried various classical machine learning models like SVM, K-NN, Desicion tree and logistic regression to get state-of-the-art accuracy.Finally I have also used deep learning models like multilayer perceptron and 1D conv-net. Uniqueness of the datset set is that it has 512 features that is the feature space has 512 dimensions.

Result of the predictive models:

Model Accuracy
logistic regression Test Accuracy: 0.9500316255534472
MLP Test accuracy: 0.932
1D Conv Test accuracy: 0.917
SVM ker = rbf Test accuracy: 0.941808981657179
SVM ker = linear Test accuracy: 0.9487666034155597
SVM ker = Polynomial Test accuracy: 0.92662871600253
KNN Test set Accuracy: 0.8548387096774194
Decision Tree DecisionTrees's Test Accuracy: 0.8358633776091081

deep-learning model

i

classical machine-leaning models

1

2

About

In this project, I explored this dataset. This is basically a classification problem. I have tried various classical machine learning models like SVM, K-NN, Decision tree, and logistic regression to get state-of-the-art accuracy. Finally, I have also used deep learning models like multilayer perceptron and 1D Conv-net

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published