solemnshark871 / churn Goto Github PK

Edit this description :)

Retrain a second SVM model having performed logistic regression

Having corrected the issue of biased data #14 we now need to retrain the SVM model to observe how it performs on a less biased representation of our dataset.

Apply K-folds CV after selecting the optimal hyperparameters

Perform PCA analysis on the dataframes (having removed persona & lamda)

As with the SVM #15 & NN #16 issues, we need to retrain our logistic regression model, originally trained by @favioacostad

Experiment with different test-train splits for the LogReg/SVM models. Try

• 80:20
• 90:10
• 95:5

Performing the model training with 250 principal components.

Review SVM model and plan how the SVM model will work for the Churn model

Alejo found an issue with our pre-processing steps where we apply a different standardisation process to each dataset (the retired vs non-retired datasets) we will need to retroactively fix this so we can retrain the models and evaluate their "real" performance.

Fixing the bias issue for the NN model (same as #15) as trained by @juanfrh7 & @favioacostad

@juanfrh7 go fill the accuracy metrics please :)

The plots for the exp log likelihood don't show and there seems to be too many of them

Exploring and analysing the data before pre-processing

7 people have 0 for lambda (after doing log likelihood MLE per age range)

Favio builds the Logistic Regression model

Lamda made our model really too good so we need to remove it

Start building a basic SVM model as a classifier (MVP)

The end date for R/NR is wrong because the window has an incorrect end date. Therefore the ages and time spent at Asw is innacurate