Page 78 - Undergraduate Research Opportunities Program Proceedings 2017-18

School of Engineering
Department of Computer Science & Engineering

This report describes the progress of our UROP - “Air Quality Data Analysis and Prediction Using Machine Learning
Approach.” In this paper we provide methods to improve the efficiency and the feasibility of predicting the rainfall at the
region for the next 1-2 hours based on the radar images. Our solution uses a two phase structure to predict the
precipitation index. Firstly, using feature engineering on the dataset to get better understanding of the global cloud
pattern, the temporal and the spatial structures and in the second stage, we feed the vectors into the Neural Network
along with the pinpoint local radar images to get the results.

A Study on the Principles and Implementations of Array Databases
Supervisor: NG Wilfred Siu Hung / CSE
Student: HYUN Jeongseok / CPEG Course: UROP1000, Summer

During the past few decades, there was a paradigm shift in the scientific research from experimental approach to data
analytic approach. The conventional DBMSs, such as Relational DBMS, is inefficient to store and process scientific data.
In order to manage the scientific data more effectively, the new concept of database come up with that selects array as
a first-class citizen. Array Database Management Systems (ADBMSs) improve the performance and functionalities from
the conventional DBMSs.This research explores the principles of array databases and conducts the experiment to
compare the performance of Array Database and relational database. The experiment was done with the datasets of
vessel traffic data, or Automatic Identification System (AIS) data collected by U.S. Coast Guard which is used for a variety
of coastal planning purposes. The process of experiment is started by pre-processing the raw data to remove the header
of csv file and remove the data of duplicated coordinates. The coordinates are consisted of lattitude, longitude and base
Date Time, thus, it is a 3-D array data. After that, the time spent for loading each csv file is recorded. Even though the
experiment origianlly purposed to compare the time spent for loading in MySQL, a famous relational DBMS, due to the
time constraint, this research is done with only tileDB, a new array database library.

Message Labelling Project for Using Machine Learning in Spammer Detection on Real-life Data

Supervisor: NG Wilfred Siu Hung / CSE
Student: CHENG Chi Fung / COMP Course: UROP1000, Summer
TANG Yubo / SENG UROP1000, Summer

This UROP is running spammer detection on a commercial dataset from Mama Helper, which is developed for employers
and HK domestic helpers’ communications. First, we contact with the Mama helpers organization and intend to help
them build a spammer detecting device. The general procedures are as follows: preprocessing, labeling, machine-
learning, detecting and what we are supposed to do is the first two part as for this UROP. First, we spent some time
learning python and relevant knowledge about machine learning and spammer detection. Then we started to write our
own code for preprocessing and labeling. The process was hard and repetitive, due to unexpected fatal error in code and
consequently rewriting the whole code. But the progress was done at the final stage. After refining the code time after
time, we got a well-functioning version and we used it to label near 10000 messages manually.

Visual Analysis of Big Data

Supervisor: QU Huamin / CSE
Student: CHIU Ka Ho / SENG Course: UROP1000, Summer

Data visualization has been prevalent around the world for years. It is extremely useful in the field of analyzing specific
objects’ changes over a period of time. In data visualization, we typically adapt tables and graphics to make data
accessible and easy to be compared. And we also gather statistics in an interval and process it to see how its

68