• Showcases CS548 Fall 2016
• Showcases CS548 Spring 2016
• Showcases CS548 Spring 2015
• Showcases CS548 Spring 2014

Sept. 21: Decision Trees

• Students: Yimin Lin, Youqiao Ma, Ran Lin, Shaoju Wu, Bhon Bunnag.

• Application Topic/Title: "Automatic selection of molecular descriptors using random forest: Application to drug discovery"

• Short description: Skillful feature selection is an essential pre-processing step for the efficient application of data mining techniques in drug discovery. This showcased paper examines a Random Forest-based approach to automatically select features for classification. The reduction of features helps to reduce the computing time over existing approaches, and permits the exploration of much larger datasets. The performance of models constructed over Random Forest-based reduced features is compared with manual feature selection when these models are trained using Random Forest, Support Vector Machines (SVM) and Artificial Neural Networks (ANN) approaches.

• Slides: CS548F17_Showcase_Decision_Trees

• Link to paper: Showcased Paper

Sept. 28: Model and Regression Trees

• Students: Nguyen Vo, Bolun Lin, Qing Xu, Ishaan Singhal, and Andi Dhroso

• Application Topic/Title: "Everyone's an Influencer: Quantifying Influence on Twitter"

• Short description:
  Quantifying how influential a user is on a social network plays an important role on online promotion. There is a large number of studies done on this research problem. One of the first attempts to measuring influence level of users on Twitter is using Regression Trees. This showcase presents how Regression Trees can be employed to derive data insights. An interesting result is that influential users are those with many followers and who have had high influence in the past.

• Slides: CS548F17_Showcase_Model_and_Regression_Trees

• Link to paper: Showcased Paper

Oct. 5: Association Rules

• Students: Jidapa Thadajarassiri, Meng Wang, Muhammed Veyis Kilincer, Sai Kiran Vadlamudi, Siqin Li

• Application Topic/Title: "Identifying combinatorial biomarkers by association rule mining in the CAMD Alzheimer's database"

• Short description:
  Association rules are simple at a conceptual level yet they can be used for much more than the market basket example that is used to introduce them. This paper uses association rules to provide an inexpensive and effective method of diagnosing Alzheimer's. The authors devise an algorithm, based on the apriori method, called SCARF which produced novel rules to better and more inexpensively detect this dangerous and debilitating disease.

• Slides: CS548F17_Showcase_Association_Rules

• Link to paper: Showcased Paper

Oct. 26: Clustering

• Students: Janvi Kothari, Sanket Gujar, Mihir Sawant, Umesh Nair, and Jin Huang.

• Application Topic/Title: "Hierarchical Aligned Cluster Analysis for Temporal Clustering of Human Motion"

• Short description:
  This paper presents an approach for clustering human motion using Hierarchical Aligned Cluster Analysis (HACA). HACA combines kernel k-means clustering with generalized dynamic time alignment kernel to cluster human motion captured in a time series.

• Slides: CS548F17_Showcase_Clustering

• Link to paper: Showcased Paper

Nov. 9: Anomaly Detection

• Students: Jun Dao, Qiming Wang, Emily Weber, Zijun Xu, Ruosi Zhang.

• Application Topic/Title: "Improved Principal Component Analysis for Anomaly Detection: Application to an Emergency Department"

• Short description:
  This paper is about applying an improved PCA algorithm using MCUSUM to detect anomalies in data collected by an emergency department. The authors combine these two techniques to better detect small anomalies in demand for patient care.

• Slides: CS548F17_Showcase_Anomaly_Detection

• Link to paper: Showcased Paper

Nov. 14: Text Mining

• Students: Zhiqi Chen, Yuchen Shen, Tianyu Wu, Di You, Xiaoyu Zheng

• Application Topic/Title: "Using text mining and sentiment analysis for online forums hotspot detection and forecast"

• Short description:
  This paper applies sentiment analysis and text mining approaches to detect and forecast the hotspot of Sina Sports Forums. First, it develops an algorithm to calculate the emotional polarity of the text. Second, K-means is used to detect hotspots and SVM is used to predict the hotspots. The performance of SVM is measured by the K-means clustering result, and the prediction result of SVM is highly consistent with the clustering result of K-means.

• Slides: CS548F17_Showcase_Text_Mining

• Link to paper: Showcased Paper

Nov. 30: Sequence Mining

• Students: Emma Clavet, Nan Hu, Shivangi Pandey, Tesia Shizume, Xiaojun Wang

• Application Topic/Title: "The Use of Sequential Pattern Mining to Predict Next Prescribed Medications"

• Short description:
  The aim of this paper is to determine whether Sequential Pattern Discovery using Equivalence Classes, a sequential pattern mining algorithm, is an effective method for identifying temporal relationships between medications and accurately predicting the next medication prescribed to a patient.

• Slides: CS548F17_Showcase_Sequence_Mining

• Link to paper: Showcased Paper

Dec. 5: Web Mining

• Students: Linh Hoang, Giorgi Gachechiladze, Tian Xie, Jiaming Di, Yang Zheng

• Application Topic/Title: "We Feel Fine and Searching the Emotional Web"

• Short description:
  This paper presents We Feel Fine, an emotional search engine and web-based artwork which utilizes web mining to collect, detect and display the Internet emotions in realtime.

• Slides: CS548F17_Showcase_Web_Mining

• Link to paper: Showcased Paper