DJEKI Essohanam

Affiliated project ACETSMIP

Affiliated CEA Université d'Abomey-Calavi - CEA-SMIA and National Open University of Nigeria - ACETEL

Title: Modeling security risks in the West African digital learning space using machine learning techniques.

Start date: 01/01/2021

Issue:
With the arrival of the COVID-19 pandemic, conventional institutions will increasingly participate in the use of the digital space for their service delivery. Hundreds of thousands of learners are opting for online programs in the West African region, resulting in high exposure to cyber-attacks. To ensure the sustainability and security of the digital learning space, it is essential to equip participants with state-of-the-art security tools and technologies. Several systems such as Moodle, Blackboard, Zoom, Coursera, edX, Google Classroom and Google Meet are used.

Digital learning systems have always been under constant pressure from cybercriminals and malware. With the increasing integration of computers, software and IT services into our daily lives, the task of ensuring data security is becoming ever more difficult. 

In a context of Big Data, where everything is connected to the Internet - from connected devices to physical and virtual terminals - and is a source of information or a potential point of attack, machine learning can help to decipher, analyze and interpret data with minimum effort. 

The simplest approach to applying machine learning to security threats is to start with a list of previously labeled anomalous or abnormal events and use a supervised machine learning algorithm to learn a relationship between the event's characteristics and this label. Compared with the explicit rules approach, many authors avoid the work of writing or rewriting rules. However, the problem of detecting threats that have never been seen before remains the same with conventional techniques. To overcome this shortcoming, some authors have adopted unsupervised ML techniques. There is a whole range of such techniques, which can be divided into two categories depending on whether or not the training set is contaminated by anomalies.