The purpose of this exercise is to predict two classifications of answer rate, Good and Improvable.
For this I was studying the data until I found those that could tell us from the previous month, what could be the metric for the next one.
- People working the next month
- New people working the next month
- Number of information workshops the next month
then i did the correlation matrix and i found that these variables has a correlation of more than 0.5, which is optimal for generating a prediction pattern
After doing the mathematical function, we can see that the accuracy in training, validation and test does not drop below 85%, which tells us that we have a fairly reliable logistic regression prediction model.
To complement the exercise, I used a neural network add-in to see if the accuracy improved using another type of prediction method, the accuracy goes up, however the disadvantage is that using the neural network we cannot see what behavior patterns it does not predict well. , since it is not a manual exercise like logistic regression.
Please find the Exercise in the link below
https://docs.google.com/spreadsheets/d/1gU1oMEy2pqsrRKU8oPasCnsxH08AU9bWcg5QVe-tXl0/edit?usp=sharing
Kaggle data set: https://www.kaggle.com/unifrancouni/call-center-metrics-dataset
Data.world data set https://data.world/axonvoltage/average-handle-time-by-day