Create a Predictive model to find users that would book an exam

Question

I'm currently working for a health company in Brazil, and I want to create a Predictive model that would find users that are one step close to book an appointment.

We currently have 1 million visits per month, and 60% of this visits are not booking anything (exam, consult, checkups, etc). I believe that with ML, we could find which users would book if we make a call or send an email.

I have been a professional web development for the past 7 years, but I don't have the skills to create a Predictive model, so I'm looking for directions, a way to start.

We do have Google Analytics, but I'm thinking to create a simple software that would track:

Age
Gender
Which pages the user visited
How many time the user expended in each page
Name

Then, I could segregate the users in two groups: the ones that booked, and the users that didn't book, and then... I'm stuck.

Could someone give a direction? Should I create a software to track this information or use GA? I want create a demo to get financing, so I don't have much time :/

Thank you.

Answer 1

Your problem is a kind of Binary Classification which predicts the probability of a target variable to be Yes(1) or No(0). But, first of all, you need some ground-truth data, because most of the state-of-the-art solutions to this problem are supervised.

If you're the admin of this system or have access to their server, you can collect these data by logging everything you need from every user and label their traces with 1 and 0 when they book or not.

After a while, you have a big labeled dataset and you can normalize and shape your inputs in a way that it fits for example to a deep neural network for prediction.

Answer 2

In addition to Moh's answer, which is impossible to disagree with, I understand you have limited time to prototype a binary classification model. Given that you don't have much experience coding machine learning models, right after you collect your data and label your dependent variable with 1 and 0, you should go for quick prototyping tools such as Orange, Weka, RapidMiner, SAS Enterprise Miner etc. These are point-and-click platforms that require little to no experience in machine learning. Very easy learning curve, you can build a proof-of-concept very easily.

Later on, depending on your needs and limitations, you can stick with these easy-to-use tools, or move to R and Python, and code your classification model.

Answer 3

It's hard to answer your question because it is too general. It's like asking how to build a website without even knowing how to do programming. But I will be constructive

Given your background and assuming you are familiar with Python, I recommend you using auto-sklearn. auto-sklearn is a library based on the popular sklearn which automates a lot of steps involved in general machine learning, like feature selection and hyperparameter-tuning. It also tries a bunch of algorithms and ensemble them to produce the best result. It is hard to beat its performs as a machine learning beginner.

While auto-sklearn is very good for a beginner to get started, I must warn you that, without fully understand the how and why it is extremely risky for you rely on it blindly. One thing of particular importance is you need to understand how to evaluate the performance of a machine learning algorithm. Please make sure you understand the difference between train, validation and test dataset, before you make any claim on how good/accurate you prediction is.