Machine learning for management decision support systems

In the last decades, data generation and collection has never stopped improving, leading to an explosion in the volume of data. Information can be extracted from such data and would be valuable for companies and institutions, especially for decision support systems. In this context, machine learning (ML) emerged.
The main goal of this thesis is to assess whether ML techniques can efficiently perform classification and which one achieves best performance. To this end, performances of ML techniques will be analyzed when facing two datasets with different characteristics, including dataset size, number of features and data type (structured vs unstructured).
The first dataset is concerned by education. The intent is to develop a new means of identifying freshmen’s profiles likely to face major difficulties to complete their first academic year. We focus on early detection using student data available at registration with a view to timely and efficient remediation and/or study reorientation. We adapt different ML methods and design algorithms to increase the accuracy of the prediction when some classes are of major interest.
The second dataset is about online reviews. In current literature, we noted significant confusion on which features and approaches achieved best performance in predicting review helpfulness. Therefore, we propose to refine the set of features that were found to be most promising in past research using feature selection mechanisms. Especially, we rely on lasso and elastic-net. Several ML methods are also compared to determine state-of-the-art methods in classifying reviews. Finally, we examine whether the classifier performance is improved by considering only a small set of high quality features, selected by shrinkage methods.