https://github.com/ekramasif/pre-processing-and-feature-engineering-with-svm

https://github.com/ekramasif/pre-processing-and-feature-engineering-with-svm

Last synced: 3 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/ekramasif/pre-processing-and-feature-engineering-with-svm
• Owner: ekramasif
• License: mit
• Created: 2021-09-19T14:56:49.000Z (over 3 years ago)
• Default Branch: main
• Last Pushed: 2022-02-13T10:34:22.000Z (over 3 years ago)
• Last Synced: 2025-01-10T10:29:52.806Z (4 months ago)
• Language: Jupyter Notebook
• Size: 3.76 MB
• Stars: 3
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Pre-Processing-and-Feature-Engineering-with-SVM

# General remarks

This assignment is meant to get your acquainted with Support Vector Machines (SVM). You

will also have to explore Pre-processing and Feature contribution quite a bit further (as

explained throughout the Lab classes). Additionally, you’re encouraged to experiment a

little further with report writing, by making your own choices and presenting it as a paper

like format if possible.

For the practical parts, you are asked to train and test a few models using SVM, and produce

what you think the best settings are for the data we’re using. Use the dataset (corona_data)

that we have used during our lab class. Note that you have to show the results based on

the big dataset. That is, you have to train your model using the “train.tsv” and the evaluation

of your final model would be based on the “test.tsv”. However, for tweaking your models,

you may use the small datasets.

# Default settings:

Run a support vector machine with a linear kernel on the multiclass classification

(cls = svm.SVC(kernel=’linear’, C=1.0). Use default settings and report results using

the portion of the test/development set. Just make sure you document what you have done

along with the results.

Reporting the Results (Evaluation Metrics):

As evaluation, you will have to report on the overall Accuracy as well as the class-wise

Precision/Recall and F1 score. Additionally, also include confusion metrics for each of your

experimentation. Please do not just put the results in the report. Try to explain a bit about

why you might thing the results have improved/degraded from the previous experiment(s).

# Output:

          

          Training the Classifier...

          SVM Accuracy =  0.4

                              precision    recall  f1-score   support

          Extremely Negative      0.479     0.240     0.320       146

          Extremely Positive      0.591     0.340     0.431       162

                    Negative      0.381     0.483     0.426       302

                     Neutral      0.453     0.414     0.433       152

                    Positive      0.324     0.424     0.367       238

                    accuracy                          0.400      1000

                   macro avg      0.446     0.380     0.396      1000

                weighted avg      0.427     0.400     0.399      1000

          Naive Bayes Accuracy =  0.269

                              precision    recall  f1-score   support

          Extremely Negative      0.296     0.199     0.238       146

          Extremely Positive      0.272     0.136     0.181       162

                    Negative      0.335     0.255     0.289       302

                     Neutral      0.246     0.474     0.324       152

                    Positive      0.232     0.290     0.257       238

                    accuracy                          0.269      1000

                   macro avg      0.276     0.271     0.258      1000

                weighted avg      0.281     0.269     0.262      1000