D208_Task2_PA

D208 - Predictive Modeling Logistic Regression Modeling

Table of Contents Part I: Research Question ......................................................................................................... 3 A. Describe Purpose of Analysis ......................................................................................................... 3 1. Summarize one research question .................................................................................................................... 3 2. Define Goals of Analysis .................................................................................................................................... 3 Part II: Method Justification ..................................................................................................... 3 B. Describe Multiple Logistic Regression Methods ............................................................................. 3 1. Summarize four assumptions of a logistic regression model ............................................................................ 3 2. Describe two benefits of using Python in support of analysis .......................................................................... 3 3. Explain why logistic regression an appropriate technique is to use based on question in Part I ..................... 3 Part III: Data Prep .................................................................................................................... 3 C. Summarize the data prep process .................................................................................................. 3 1. Describe Data cleaning goals – See attached .ipynb File ................................................................................... 3 2. Describe dependent and all independent variables ......................................................................................... 5 3. Generate univariate and bivariate visualizations of the distributions – independent and dependent variables, include dependent variable in bivariate visualization .......................................................................... 7 4. Describe data transformation goals that algin with your research question and the steps used to transform the data to achieve goals, include annotated code ............................................................................................ 13 5. Provide Prepared set as a CSV file ................................................................................................................... 14 Part IV: Model Comparison & Analysis ................................................................................... 15 D. Compare initial and reduced linear regression model .................................................................. 15 1. Initial multiple linear regression model with all variables from part C2 ......................................................... 15 2. Justify statistically based feature selection ..................................................................................................... 16 3. Provide reduced linear regression model ....................................................................................................... 19 E ....................................................................................................................................................... 20 1. Model Evaluation Mettric explanation ............................................................................................................ 20 2. Confusion Matrix & Accuracy Calculation ....................................................................................................... 21 3. Attached code .............................................................................................................................. 22 F. Summary ...................................................................................................................................... 23 1. Discuss results ................................................................................................................................................. 23 2. Recommend course of action .......................................................................................................................... 24 G ...................................................................................................................................................... 24 H ...................................................................................................................................................... 24 I ........................................................................................................................................................ 24

I first began by identifying that there were null fields in InternetService. Using similar techniques that I used in D206 I filled those nan fields by using the .fillna() method. It’s an easy way to fill null fields without compromising the data or removing excessive rows of data from the data source. Once the nulls were taken care of, I was able to identify outliers in the data by using box plots. It did appear that there are multiple fields with outliers: Income, Children, and Outage_sec_perweek. I determined to identify the count of outliers, and then replace them if they were several standard deviations above the mean. The threshold set was 3. The outliers outside of three standard deviations were replaced using z-scores. With the null values and outliers taken care of, the data cleaning step was complete.

2. Describe dependent and all independent variables The original data set contains 50 columns and 10,000 rows of customers. For my analysis I will be focusing on ‘Churn’ as my dependent variable. I will also retain a summary describing ‘Income’, ‘Outage_sec_perweek’, ‘Tenure’, ‘MonthlyCharge’, and ‘Bandwidth_GB_Year’. These are my nominal values. The categorical values that will be retained for independent variables are ‘Area’, ‘Contract’, and ‘PaymentMethod’. The dependent variable, Churn is binary of ‘Yes’ or ‘No’. ‘Yes’ indicates that the customer churned as a customer and ‘No indicates that they did not. ‘No’ is the top response with 7350 and 2650 were ‘Yes’. The independent variables that are nominal or continuous are Income which has 10000 rows, a minimum stated income of $348 and maximum of $258900. The average income per year for the customers is $39806. Outage_sec_perweek is also a nominal value with the average of 10 seconds per week, max of 21 and minimum of .09 seconds per week. Tenure is the lengt of time a customer has been a customer with the average of 34 months with the provider and max of 71 months. Bandwidth_GB_Year is a nominal independent variable and has an average of 3392 GB per year, maximum of 7158 GB per year and minimum of 155 GB per year. MonthlyCharge is the final nominal independent variable and has an average of $172.62, the maximum payment per month is $290.16 and minimum is $79.98. There are 4 categorical independent variables, and they are Area, which is the type of area a customer lives in. The top area for our customers is Suburban with 33.46% of customers, 33.27% Urban and 33.27% Rural. Contract is the type of contract our customers are on. The most frequent contract is month-to-month with 54.56% of customers, Two Year at 24.42% of customers and 21.02% customers at One year. Finally, the type of payment method is the last independent categorical value. The top payment method is Electronic Check with 33.98% of customers utilizing e-checks, 22.9% mailing in checks, 22.29% having a bank automatic transfer and 20.83% having automatic credit card transactions. Churn is the final categorical of Yes or No. And 73.5% of the 10,000 rows are still customers.

Now the goal is to determine what aspects of these independent variables influence Churn. 3. Generate univariate and bivariate visualizations of the distributions – independent and dependent variables, include dependent variable in bivariate visualization Dependent variable univariate visualization for monthly charge:

Independent univariate continuous variables visualizations: Independent univariate categorical variables visualizations:

Bivariate visualizations:

4. Describe data transformation goals that algin with your research question and the steps used to transform the data to achieve goals, include annotated code The data transformation that I performed was one-hot coding. The fields that were affected were the categorical fields Churn, Contract, Payment Method, and Area. Churn is the dependent field and the others are the independent categorical fields I chose to test regression with my dependent variable. The reason these had to be transformed using one-hot coding was because they were not nominal fields that could be ranked using a ranking system. They also couldn’t be converted to True or False. One-hot coding takes a column and pulls out the responses to create a separate column for each possible value and assigns a 1 or 0 depending on what the row contained. This will assist in down the line linear regression. To help mitigate multicollinearity the first column is dropped. Unlike in the linear regression model, Churn is our dependent variable for logistic regression and therefore will not need to be transformed into a numerical, binary field. 5. Provide Prepared set as a CSV file. See attached C5_preppeddata.csv

2. Justify statistically based feature selection Using the VIF score, it indicated there was multicollinearity for Tenure and Bandwidth_GB_Year. The threshold is 10. Once removed, Bandwidth_GB_Year decreased below the threshold and I ended up leaving that in. Once multicollinearity was analyzed, I used p-values to determine the independent values to retain:

Once those values were removed I wanted to check VIF again just to verify it was not affected once the other fields were removed. Checking VIF (multicollinearity) again:

3. Provide reduced linear regression model

E. 1. Model Evaluation Mettric explanation Initial Model: Reduced Model: