Research

Usage Based Emotion Detection and User Identification (January 2015 to January 2017)

We are working on detecting user’s emotion based on keystroke dynamics and mouse usage pattern. Emotion is, perhaps, the most critical attribute of living beings that is extremely difficult to detect and generate artificially. Although there exist several approaches for detecting emotion, most of the approaches require hardware overheads (body sensors, camera etc.). We tried to find a simpler way of detecting emotion using keyboard and mouse usage data only. We have used several existing classifiers (KNN, KStar, RandomCommittee, and RandomForest) and a newly proposed classifier namely Bounded K-means Clustering, to analyze those usage data for different emotional states. I worked under the supervision of professor Dr. A.B.M. Alim Al Islam. This work appeared in Proceedings of 2017 International Conference on Networking, Systems and Security (NSysS), IEEE, Dhaka, Bangladesh, 2017. (pdf)

VidSplit: An Approach of Exploring Active Learning from Video Tutorials using Online Learners (September 2017 to present)

Video tutorials play a significant role in learning feature-rich software. However, sometimes conventional video tutorials become lengthy and tedious to follow. This is due to the fact that, tutorials are not designed in a way so that learners can interact and participate actively. Therefore, they become unconscious and eventually overwhelmed with the information provided by the tutorials. In this work, we present VidSplit, an online video tutorial watching platform which prompts questions to the learners while they are watching the video. Learners interact with VidSplit by giving yes or no feedback and learn from the interaction. We performed a study with 8 participants and our findings suggest that learners learned more in less time from our approach than any other previous or conventional approaches. Our findings also point to a future scope where interactive tutorials can be designed to support active learning and to develop workflow.

On-Line Inventory Management Algorithm Considering Expiry Date of Products (September 2017 to present)

Inventory management problem, where the customer number is unknown to the inventory owner has been studied in the literature. However, to the best of our knowledge, there has been no study which considered the expiry date of the products. In this study, we investigated the online inventory management problem considering the validity of a product from 1 days to n days. Here, we have come up with an online algorithm where products have a validity of 1 day and 2 days. We have also shown how the advice of different sizes can improve the competitive ratio of our algorithm where products have a validity of 1 day.

A Comparison Study of Canonical Correlation Methods on Multi-omics Data

Canonical correlation analysis (CCA)-based methods are commonly used for reducing the dimensions of high-dimensional multi-view (multi-omics) datasets to analyze the associations among the features from different views and to make them suitable for downstream analyses (classification, clustering etc.). However, there is no well explored comparison study for CCA methods with application to multi-omics datasets (specially microbiome and gene expression datasets). In this study, we address this gap by providing a detail comparison study of three popular CCA approaches: regularized canonical correlation analysis (RCC), deep canonical correlation analysis (DCCA), and sparse canonical correlation analysis (SCCA) using a multi-omics dataset consisting of microbiome and gene expression profiles. We evaluated the methods in terms of the total correlation score, and the classification performance. We found that the SCCA provides reasonable correlation scores in the reduced space, enables interpretability, and also provides the best classification performance among the three methods.