Competitive Machine Learning

Practical course in the Winter Semester 2015/2016. Prof. Tobias Scheffer.

Course Info

• Weekly discussion every Monday at 13h00 in Seminar Room of the Informatics department.

The course is aimed at students who have previous experience in the field of machine learning, such as those described in the "Machine Learning", "Machine Learning 2", or "Intelligent Data Analysis" courses.

Project Description

The project class focuses on the problems of identifying fraudulent online loan applications and of predicting the default risk of online loan applications. You will develop and study methods that exploit globally available geographical data (such as Google Maps) in order to discover wealth indicators. You will evaluate the method by its ability to either predict if a credit is regular/fraudlent or predict the risk of paying back the entire amount on time/defaulting, using available data.

The emphasis will be put into using features that can be extracted by using the available features. Your job will be to build and evaluate these features regarding their relevance to either the credit assignment/fraud detection projects. Examples of features can be for instance, the user’s browser fingerprint, geographical location, operating systems, types of devices, and even the local time of day at which a credit is applied for.

Evaluation

In this class, you will examine, in consultation with your supervisor you will first choose a concrete modeling problem: Predict Fraudulent/Regular behaviour, or, Predict risk of defaulting. We will develop and implement appropriate procedures of machine learning, and evaluate them using available data.

Evaluation will be done by written report and presentation of the developed solution.

Data

Data will be sent by e-mail to the interested students.