Which Machine Learning technique should i use for this kind of Prediction?

which Machine Learning Technique should i use to create a Model to Predict customers who are able to upgrade from basic phone to smartphone ?

I have a Dataset which contains information about customers like Age, Address, Location and Consumption of Calls and Data .

are you familiar with Sklearn?
– Gabriel M
Jun 30 at 6:37

Can you provide more details of your data. Do you have target labels (is it supervised or unsupervised)? What is the distribution of your target variable (say x% of people in your data converted to smartphone and (1-x)% did not.) ? Perform some data analysis to choose your ML algorithm.
– Ankita Mehta
Jun 30 at 6:56

the dataset is about 10 Millions and i have a supervised target and the distribution of my target variable is to say if this person will upgrade or not
– Anis Amh
2 days ago

4 Answers
4

It depends on several factors:
Size of dataset: small datasets are not well suited to deep learning, but can be exploited whit other machine learning techniques like SVM,ILP or C4.5 trees.

Categorical versus continuous variables: SVM for continuous, ILP for categorical, C4.5 for both. I don't know for deep neural networks

White-box versus black box: Do you want to know the reasons behind your predictions? ILP or C.4. It doesn't matter? SVM or Deep.

Of course there are other ML learning techniques that you can try such as Bayesian Networks etc..

Try Weka package ;)

In yout case i will use C4.5 trees

Try for tensor flow which support neural network classification algorithm based text classification features and number classification features. The target output class will be positive and negative.

Refer to the following link from tensor flow for text classification

Try GBM and SVM first.

You may want to create additional features first. Below are some examples:

Consumption_this_month,
Consumption_prev_month,
Consumption_2_month_ago,
Consumption_3_month_ago,
...

Age_group,
Prefix_of_Zip_code

...

It's all feature engineering after all.

The best classifiers in terms of ease of use and effectiveness are the ones implemented in sklearn.

I'd recommend Gradient Boosting Classifier
or SVM
or Decision Trees

The only problem that you may face is introducing the Location/Address information.
This may require some feature engineering. Like implementing boolean features to encode the location in terms of country or city for example.

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.