Dat690 Final Presentation

Welcome to my G. E.

employee attrition proposal

for data 6 90

by Christopher David.

First we start off with the introduction

which continues the problem,

the background,

the methodology

and the resources.

The background is geez employee

attrition rate has increased

to hire and train a new employee.

It cost 80%

of the person's salary.

We have been asked to curate a model to predict

reasons for employee Trish situation.

G. Has highly trained employees

that are hard to replace.

This is why the increased cost of 80% of the person's salary

to replace an employee.

They are multifaceted business

that runs different avenues

of revenue streams

that's going on to the problem.

The problem is he has seen an increase in its employee attrition

which leads to the following impacts,

loss of experience employees.

The increased cost of training new employees

and time to train your employees causing loss of throughput

the methodology

in the methodology, we have the model type,

the method

and an example of a flow chart we're using

first we go to the model

model type logistic regression model is a model type. We use.

This model

is considered as one of the industry standards from pediatrician

and the results can be easily to explain

other models are available like the Random forest model

but can get very difficult

with more variables that are added.

Also logistic regression works well

when you have a binary system

of zeros and ones which are attrition was either yes or no.

Next is our method.

The C. R. S. I. P slash DM method

was followed while creating this model. This is an industry standard

to run models

professionally.

The next is an example of a flow chart.

So the trucks were created to ensure that

there is a consistent process that was followed

and data cleaning, data appropriation and data delivery.

Our presentation

moving away from the methodology,

we next moved to resources.

The reason resources to run this model,

our HR data and employee resource to run the model,

the HR data. We use a report that's already been created

and distorted our HR warehouse.

This report is updated each night

and the report has the required information so no

other reports would be needed to be generated.

Next employee resources

Reimported resources to run this model, it takes about one person.

It takes about 3-5 seconds to run

depending on the data size.

No other

employees or other systems will be needed to run this.

Besides our studio

moving away from resources

that concludes our introduction.

our pilot plan.

We use the current data provided by G. E H. R.

To build a model plan

The requirements to build a model

where 70% accuracy or better.

First we go into the results explained

the results explain the pilot was his success.

It was 88% accurate and was able to find the top reasons

why employees choose to leave ge

moving the results explained. We move into the results itself.

The table below shows the most significant reasons on why people leave G

as you can see.

There are some that we are directly responsible for

and some

that we are not directly responsible for.

One that we are directly responsible for is how often our employee travels.

And you see at the list

frequently seems to have a higher attrition rate.

Then

not traveling at all.

1/2 that we don't have direct influence our

the distance from home.

How far are employees leave from home?

So a balance should be created

when changing HR guidelines are adapting guidelines

to the model results.

The next is you can see the ROC curve

in a rock curve. The area under the Roc curve was

0.8684,

Anything over .5 shows that the model is a good fit

And as you can see, we're at .8

moving

from directive

and we then moved to the confusion matrix.

If you should make sure it tells us how accurate we are.

And as you can see, we are 88% accurate in this model

For a pilot phase

and test you know, building our model 88% is a good rate,

which means the employee the GE's requirement

For over 70

moving from the pilot plant.

We then move into

our testing phase.

Since our pilot was successful, we decided not to modify our model,

they go straight to testing,

which means we were using all the variables

that were provided by the HR spreadsheet.

Again, we go into the results, explained

results of Swain

we we tested our model and found it was equally as accurate.

Is there a testing model?

The results

show the same results as we got under testing model

and again

some of the variables are under control and

some of the variables are not under control.

And for this under our control

we need to look at the job sales

representative or representatives of sales

because that has a higher attrition rate than other types

of job roles that we have at G. E.

one that is not under control

is the total working years.

We cannot, you know,

in our control is not how many hours or years our employees work

moving. Next is the roc curves.

You can see a rocker did drip a little at 0.8178

But it is still within the .70 that we have been

asked by ge to make sure our model is accurate.

Our confusion Matrix

did say the same.

Thanks.

The surrounding was a little higher at 89% accurate.

So this model is a good fit for G E s employee attrition.

Moving from the testing phase,

we then

we'll move into

the stakeholders.

So I G which is a multifaceted company. There are a lot of stakeholders.

First is our Ge managers.

Team managers need the output of the model in order to have steak conversations

with their employees.

The output will give them topics to inquire

about within with their employees.

So

what does that mean by state conversations. The state conversation is

when you talk to your employee kind of as a sales representative

going, I understand that

over time is an issue or travels an issue.

They can talk to the employee to understand how they can get

together and make that employee's job involvement and job connection to G.

E. S. Plan. Better.

This will help our employees that want to stay

and maybe even recruit other employees.

The next is the HR department.

So the HR Department has two different facets.

The HR will use the output from the

model to determine what guidelines would be changed.

Student employees leaving and in some cases like

ge the HR can do the hiring which would

mean they have to also find higher and

understand how to train new employees for the business

moving from the HR department we have the I. T. Department.

The I. T. Department not only needs to know how to run the model

but they also need to know how long it takes the model to run.

Another thing that the I. T. Department is responsible for is making sure that our I.

T. Systems can handle the amount of people working and also deactivate and activate

employees accounts.

Then we move into an employee themselves.

The employees are the most important that would benefit from this model

because of the whole purposes model

is to entice employees to stay.

So not only does the employee when they leave give us the data of why they left,

but we also need to be seeing how we keep our employees with this model

in this model.

Once we adapt the guidelines

will help us entice our employees to stay at ge

moving away from the stakeholders.

We then move into

our invitation plan

which we have modification rules.

Results, explain results

ROC curve and are confusing matrix.

They don't modifications. Again, we did not modify the data set

for the verification data set or any future data sets,

the HR model or the HR spreadsheet that is

record it every night

and will be supplied to us.

Can be directly entered into our our studio model

and be able to be used to work on the model.

The results explain what the verification data

was. A model is trying to be more accurate than during the testing phase.

The results again fruit at the same results

as both during the testing and the training

and as you can see

right, we have the highest significant value on all values here.

And these would be given to g as the top reasons why

employees leave so that they can figure out or adapt their guidelines

and how to entice their employees to stay,

especially ones that fit in multiple of these categories.

The ROC curve.

There's .8619 showing this model is a good fit

again over the .70 requirement

for the 72% requirement

From ge we are at 86

And the confusion makes it sort of verification data

as you can see we were 92.5% accurate.

This number is well above what G. E. Is asking for

as well as our model is able to predict

All but 15 times

Out of the 200

Line items that were given us. So 200 employees that were given us

this model is very accurate in helping ge to determine to retrain their employees.

Next we go into

our summary or our conclusions.

The budget for the new data sets and the recommendations.

First we're going to our budget.

We know

with the knowledge of Y G. E. Employees are leaving

it's hard to modify guidelines to retrain included employees.

This reduced the new higher budget allocation

by keeping our employees and having to stay conversations

and having guidelines that are most up to date

where our employees are showing more jobs involvement and more job connection.

The ease budget to hire new employees will greatly reduce

G will be able to focus more on increasing

their product to me and developing new ideas

than worrying about hiring new employees.

The new data sets.

The new data sets

can be easily used

right with the R. Q. R. Code filing

sets that have been entered into the model.

It is

the r script that is sent with this model is open

to a new data set it just needs to be entered.

Will also be showing I. T. Department and I. T.

Representatives on how to add data sets to the model.

To make it easy for future data sets to be entered

recommendations.

The model should be run once a month to see changes

and why employees are leaving.

The reason why the once a month was set

was to ensure that any guideline changes by G by G E H R.

Had time to filter through the employees

changing a guideline today. You will not see the effect tomorrow.

We need to allow time for the guideline change

to filter through employees.

We also suggest

that a survey should be created for real time data on how guideline changes affect

employees.

This will allow G E G E H R. To see the change affected

immediately.

This survey can be handed to all employees

and asked,

we are about to change this guideline. How would you feel if we did a B or C.

Another into another

survey that could be created

to help this model is an exit interview

while HR does capture all this information.

The employees should be allowed time

too right

how they feel and why are they leaving the company?

Yes this will make it a little harder for G E H.

R to sort through so many different writings

but it does allow the employee to free flow and give

us information that may not be captured in a closed,

open, closed system

or a close ended survey

at the.

Dat690 Final Presentation

Chapters

Audio TranscriptAuto-generated