Group05 Dashboard

Visual Application for Time Series Clustering

Project Proposal

Data Preparation & Dashboard Design

Poster

Final Report

Application

Data Source

For this project, Citibike is the data set used for development of app. Data was obtain from Citibike System Data and a subset of the first week of August 2018 (01 August - 07 August 2018) is used.

Field Name	Description
tripduration	Duration of the Trip
starttime	Start Time of the trip
stoptime	End Time of the trip
start_station_id	ID of the Start Station
start_station_name	Name of Start Station
start_station_latitude	Latitude of Start Station
start_station_longitude	Longitude of Start Station
end_station_id	ID of End Station
end_station_name	Name of End Station
end_station_latitude	Latitude of End Station
end_station_longitude	Longitude of End Station
bikeid	ID of bike
usertype	Type of User (Customer = 24-hour pass or 3-day pass user; Subscriber = Annual Member)
birth_year	Year of Birth of Subscriber
gender	Gender (Zero=unknown; 1=male; 2=female)

Dashboard Design

This project aims to provide an interface for user to apply time series clustering to time related data so that they can perform clustering analysis without the need to code and visualise the result in a more interactive and visual manner. However, the dtwclust package output plots are based on default R base which can be further improved in terms of visualization.

Design Specification to Improve

The dashboard aims to bring about the following improvement of the current packages:

• Static Visualizations

The visualizations provided in current dtwclust packages are static, where users can plot dendrogram, series, centroid or sc (series and centroid) to visualize the time series clustering. However, the plot is static which is difficult for users to do identification of the cluster. For example, user is not able to identify the variable of selection within the cluster.

• Manual Calibration

User are only able to manually calibrate the key parameters such as, type of clustering, distance algorithm, centroid algorithm, number of cluster and method of agglomeration (for hierarchical clustering).

Choice of Visualization and Critic

Critics on the default visualizations provided in the dtwclust packages will be discussed as well to the areas for improvement for our visualization designs

Visualization	Discussion
	The current dtwclust package plot of series and centroid allows the user to visualize the time series of different cluster. The different colour lines represent the individual series of the variable (label). However, the plot only visualizes the time series of different cluster but user is not able to determine which series belongs to which cluster.
	The current cluster dendrogram plot does not help in visualising the cluster especially when the dataset is large. Hence, it is difficult for user to visualize any of the cluster output. The series is all cluttered together and not aligned which results in difficulty in visualizing the dendrogram plot.

Functional Design Specification

The project also aims to incorporate the following functional design in the application to enhance user experience and create better engagement.

Storyboard	Detailed Design
	User Friendly Data Preparation The application will enable the user to import the dataset and a data table will be generated to allow the user to view the dataset of choice. From there, the user can select the data column of choice to be analyse. The data table also allows user to search within the dataset.
	Visually Driven Data Exploratory As one of the important data preparation steps in any time series analysis is to be able to aggregate the time, the application will have an added feature to enable user to do so with ease. In addition, the user is able to visualize different time aggregation of their dataset. The application will have two different kind of data exploratory, where one is the timeline of the dataset, which will be a normal line graph. If the dataset contains longitude and latitude data, the user can also opt to explore the data using geospatial.
	Data Analytics and Cluster Evaluation The data mining technique of interest is time series clustering. Clustering technique itself have various types, different distance and centroid algorithm as well as different method algorithm. The application main function is to allow user to explore different clustering technique on time series dataset. With that, the main focus of the data analytics tab is to allow user to do so without the need to coding, but a simple user input interface. The application aims to expose the Cluster Evaluation to user. In addition, to enhance the experience by recommending the number of cluster based on cluster evaluation output. User have the flexibility to use the recommended cluster or desired number of cluster. The corresponding model will be retrieve from the models computed for subsequent analysis.
	Diving Deep into Cluster Characteristics Finally, with the time series clustering performed, the application also allows user to further analyse the clusters. User are also able to select different variable from the dataset to investigate the characteristics and the idiosyncrasies of each cluster, for example the distribution of gender or age within the cluster.