Sei sulla pagina 1di 4

AI ML TELECOM PRESENTATION.

txt
MOBILE TELECOM BUSINESS

MACHINE LEARNING TELECOM

? ITS ARTIFICIAL INTELLIGENCE APPLIED TO TELECOM


SERVICES

BEFORE SPEAKING ABOUT

? IN OUR CASE WE ARE GOING TO SPEAK ABOUT CUSTOMER


CHURN PREDICTION

IN FACT TELECOM CHURN MEANS THE LOSS OF CUSTOMERS


OR CLIENTS

? TELECOM COMPAGNIES OVER THE WORLD LOOSE YEARLY A


POURCENTAGE OF

THEIR SUBSCRIBERS THAT SWITCH TO ANOTHER OPERATOR

IN OUR CASE WE APPLY ARTIFICIAL INTELLIGENCE TO A


99999 SUBSCRIBERS

AND FOCUS ON THE METRICS THAT HELPS TO PREDICT CHURN


IN ORDER TO MINIMIZE

THE CHURN .

With millions of subscribers, a variety of new


products,

Pge p
AI ML TELECOM PRESENTATION.txt
bundled and customized solutions, the operational
support
services like service configuration, order
fulfillment,
customer care, and billing are becoming increasingly
complex. Hence,
the cost of handling these operations require
resources and different
tools, thus, increasing the financial overhead.
Telecommunication provito upgrade their IT and
connectivity infrastructure
and focus on providing data and voice services that
are high quality, reliable,
and affordable. Security of the networks has become
a major priority for the telcos
and they are facing challenges with the emergence of
new threats that are powered by
new technologies. So, a number of operational and
technical innova

high value customers


GOOD PHASE

1 -> 7 Import of librairies ,algorithms


,metrics ..that we will use after
And info about Dataset

8 -> 14 Exploratory Data Analysis And


preprocessing
checking that numbers are unique or not

Pge p
AI ML TELECOM PRESENTATION.txt
(so 99999 different subscribers)
checking for missing data over >= 70
per cent
Creating new features that telecom
operators uses called high values customers
(the good phase month 6 and month 7)
Total recharge amount logic = Total
recharge data + Total recharge amount
so we have a new dataset df1 with high
values customers.

15 -> 21 we drop not needed data


and create a new important feature the
churn with values 0 or 1
the dataset df1 had changed since we
have dropped columns that create new features

22 -> 26 following the preprocessing


another cleaning to drop what not
necessary in our tretament
remplacing the rows more than 55 per
cent witn nan

27 -> 49 data visualization


50 -> 70 data to numbers
following preprocessing
creating new feature the factor
retaing the customer

71-> 101 Apply models tree decision


/regressions /KNN classifier
visualizations
model performance metrics

Pge p
AI ML TELECOM PRESENTATION.txt

5. Model Building
5.1. Baseline Model
5.2. Synthetic Minority Oversampling
TEchnique (SMOTE)
5.3. Recursive Feature Elimination
5.4. Univariate Selection
5.5. Decision Tree Visualization
5.6. KNN Classifier
5.7. Vizualising a decision tree from random
forest classifier
5.8. A random forest classifier.
5.9. Gaussian Naive Bayes
5.10. Support Vector Machine
5.11. Tuning parameters for support vector
machine
5.12. LightGBMClassifier
5.13. XGBoost Classifier
6. Model Performances
6.1. model performance metrics
6.2. Compare model metrics
6.3. Confusion matrices for models
6.4. ROC - Curves for models
6.5. Precision recall curves

Pge p

Potrebbero piacerti anche