(Day 11) Joined a 'bank churn prediction' competition on Kaggle

Ivan Ivanov · January 12, 2024

Hello :) Today is Day 11!

A quick summary of today:

  • Binary Classification with a Bank Churn Dataset on Kaggle
  • My notebook with various SVM models for bank churn classification

Data

image

I tried various SVM models and SVM-hybrind models, and below are the results from all

Basic SVM model

  • Accuracy: 0.8554246069015663
  • Precision: 0.7598201609086607
  • Recall: 0.4608208955223881
  • F1-Score: 0.5737001965338574
  • AUC-ROC: 0.7109204519856266
  • Confusion Matrix:
  • [[25024 1015] [ 3757 3211]]

SVM and RandomForestClassifier model

  • Accuracy: 0.7972854243039356
  • Precision: 0.5216507738002188
  • Recall: 0.4789035591274397
  • F1-Score: 0.49936401047512163
  • AUC-ROC: 0.6806937627427974
  • Confusion Matrix:
  • [[22979 3060] [ 3631 3337]]

SVM and KNeighborsClassifier model

  • Accuracy: 0.8378828733298997
  • Precision: 0.6524608712049783
  • Recall: 0.4965556831228473
  • F1-Score: 0.5639312199494744
  • AUC-ROC: 0.7128886177049008
  • Confusion Matrix:
  • [[24196 1843] [ 3508 3460]]

Hyperparam tuned SVM model I did manual hyperparam tuning to see the effects of various changes, and this took me the most time, and in the end - it had the same result as the base model

  • Accuracy: 0.8554246069015663
  • Precision: 0.7598201609086607
  • Recall: 0.4608208955223881
  • F1-Score: 0.5737001965338574
  • AUC-ROC: 0.7109204519856266
  • Confusion Matrix:
  • [[25024 1015] [ 3757 3211]]

Overall, the training time for each model was very long. I guess SVM is like that.

That is all for today!

See you tomorrow :)

Original post in Korean