Our department attaches great importance to the research training of students, and most courses have projects. Here are some research projects and course projects done in the previous year.

Note: Some of project reports or slides are in Chinese

Psychology

Impact of Attention on Unconscious Perception and Its Stages of Action

Reviewed literature on attention, consciousness, and unconsciousness, and designed experiments on unconscious perception. Independently conducted the experiments and developed a MATLAB program to ensure scientific rigor. Ultimately, I analyzed the data using SPSS and was responsible for creating clear charts, following relevant guidelines to write a research paper of over 10,000 words.

see: Paper

Multimodal Data-Driven Personalized Music Therapy Research

PI: Prof. Kejun Zhang , Department of Computer Science

In this project, I integrated neuroscience and artificial intelligence to create a database of multimodal physiological signals from patients undergoing music stimulation. I explored the mechanisms behind music therapy and developed a personalized, multimodal data-driven music therapy model, enhancing the quantification of its effects. Additionally, I designed a stress relief intervention system that utilizes brainwave (EEG) and facial control in music therapy, selecting appropriate scales and conducting experiments to validate its effectiveness.

Personality Prediction from MOBA Gameplay Behavior

PI: Prof. Shubin Yu，Department of Communication and Culture，BI Norwegian Business School

In our research, we applied supervised machine learning models to analyze the preferred and least preferred roles and skills of 1,058 employees, aiming to predict their personality traits. We split the dataset into a 75/25 training-test set, ensuring robust evaluation through a 10-fold grid search for hyperparameter tuning. This approach led to promising results, with our model achieving a macro F1 score of 0.75 for neuroticism and 0.66 for extraversion, highlighting its effectiveness in capturing these personality dimensions. Currently, we are revising a paper based on this study, which we plan to submit to Computers in Human Behavior.

See: Manuscript

Human-Computer Interaction

Research on Situation Awareness in Human-Autonomous Agent Interaction

PI: Prof. Zaifeng Gao , Department of Psychology and Behavioral Sciences

In this project, I conducted a competitive analysis and generated various camera perspective scenarios for both human-driven and autonomous driving situations. I reviewed relevant literature on autonomous driving and designed experiments to analyze the dependent variables associated with different camera perspectives. Additionally, I developed a processing program for an Autonomous Driving System (ADS) intent expression questionnaire and utilized Python for data analysis and scale calculations, contributing to the overall understanding of camera perspectives in driving contexts.

Marketing

Computer Sales Forecasting Based on Machine Learning

PI: Prof. Weihua Zhou , Department of Data Science and Management Engineering

In this project, I gained valuable experience in applying machine learning techniques to real-world sales forecasting challenges. Specifically, I honed my skills in feature selection using the Boruta algorithm, enabling me to identify key factors influencing sales performance effectively. The use of the XGBoost incremental algorithm enhanced my understanding of model optimization and evaluation, as I achieved a commendable forecast accuracy of 78%.

Additionally, compiling a comprehensive rental-return analysis report strengthened my ability to translate data-driven insights into actionable recommendations. This experience has deepened my analytical skills and equipped me with practical knowledge in data science and machine learning, which I can apply in future projects and research endeavors.

Date Science

Facial Palsy Detection based on Yolov5

PI: Prof. Edgar Lobaton , Department of Electrical and Computer Engineering

Facial palsy, affecting 1 in 5,000 people annually, causes sudden loss of muscle control on one side of the face, resulting in asymmetry. While the exact causes remain unclear, infections are key contributors, with early detection critical, especially for high-risk groups like those with diabetes or HIV. Traditionally, diagnosis involves invasive methods like EMG and ENoG. This project aims to leverage AI and computer vision for non-invasive early detection using images and videos. I analyzed the YFP dataset (32 videos, 21 patients) and implemented YOLO-based face detection in PyTorch, optimizing models through data augmentation and hyperparameter tuning. After 50 epochs, I achieved 71.62% accuracy and 86.83% recall. Performance was monitored via bounding box, objectness, classification losses, and evaluated with precision, recall, and F1 scores. I focused on detecting paralysis in the ocular and oral regions using sequential test image analysis.

see: Poster

Energy Data Analysis

The project is the coursework of Statistics and Big Data Analysis. This project investigates energy consumption patterns to understand their relationship with economic development and to predict future energy demands and structures, particularly in China. We aim to reveal how energy use influences economic growth and provide insights for sustainable development. Key components of the project include data import, preliminary analysis, parameter estimation, hypothesis testing, regression analysis, and machine learning. The findings will offer valuable guidance for energy planning and sustainability.

see: Slides

Optimized Multi-beam Bathymetry Model Using Genetic Algorithms and a Greedy Strategy

I participated in the National Mathematical Contest in Modeling 2023, where I applied mathematical modeling techniques to solve complex, real-world problems as part of a collaborative team.

The goal is to enhance the efficiency of ocean floor mapping by minimizing the total length of survey lines while ensuring complete coverage. For sloped terrains, we calculate the optimal overlap between adjacent measurement strips and determine the shortest survey line configuration needed for full coverage. In cases where the terrain is uneven, we apply a greedy strategy to balance between minimizing the length of the survey lines and reducing unmeasured areas. We use a genetic algorithm to optimize the angle and placement of the survey lines, achieving significant improvements in measurement efficiency. Through this approach, we successfully reduced the overall line length, ensuring accurate bathymetric data collection with minimal redundancy. The results of our model can be used to improve ocean mapping accuracy and efficiency for real-world applications.

This work improved my modeling skills (Python) and practiced my drawing skills(flowchart&statistical graph).

see: Paper

Swings in Tennis Match: The Impact of Momentum on Flow of Play

I participated in the Mathematical Contest in Modeling (MCM/ICM) 2024 and this work won the Honorable Winner Prize.

My work primarily involves analyzing match data to study the impact of momentum shifts on the flow of tennis matches. I developed a model based on qualitative and quantitative indicators to quantify momentum and analyzed the relationship between momentum, swings in play, and runs of success. Finally, using logistic regression and time series analysis, I achieved high accuracy in predicting key swings in matches and provided strategic recommendations to coaches for handling momentum changes.

This work improved my modeling skills(Python) and practiced my writing(Latex).

see: Paper