https://github.com/smabdullah/sound-pressure-level-prediction
A regression model to predict the sound pressure level
https://github.com/smabdullah/sound-pressure-level-prediction
bokehplots regression-analysis regression-models
Last synced: 6 months ago
JSON representation
A regression model to predict the sound pressure level
- Host: GitHub
- URL: https://github.com/smabdullah/sound-pressure-level-prediction
- Owner: smabdullah
- Created: 2019-03-13T22:55:41.000Z (over 7 years ago)
- Default Branch: master
- Last Pushed: 2019-03-21T00:23:52.000Z (over 7 years ago)
- Last Synced: 2025-01-22T05:15:34.067Z (over 1 year ago)
- Topics: bokehplots, regression-analysis, regression-models
- Language: HTML
- Homepage:
- Size: 731 KB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Sound pressure level prediction
This project works with the NASA dataset which contains various wind tunnel speeds and angles of attack. It predicts the sound pressure level.
This dataset is collected from https://archive.ics.uci.edu/ml/datasets/Airfoil+Self-Noise and known as Airfoil dataset. It has five feature attributes, all are numeric and one predicted output, the sound pressure level.
This is a regression problem. We start with the linear regression (OLS) and then try the decisiontree and random forest.
The accuracies of using different matrics are presented in the following table.
**Model** | **R^2** | **MSE**
-------------------|----------------|----------------
Linear (OLS) | 0.5585 | 20.7651
Decision tree | 0.8909 | 5.1321
Random forest | 0.9279 | 3.3899
The visualisation of the generated decision tree can be found in the **decision-tree.pdf** file
The current implementation does not consider k-fold cross validation and (hyper) parameter tunning. In the next run, I will use grid search to do cross validation and paramter tunning.
This particular dataset has five independent variables (feature matrix). We can use PCA (unsupervised) or LDA (supervised) to extract the most valuable features to visualise the regression outcome.