School of Infocomm (SOI)

End-To-End Implementation of Tele-Imaging, Video Consultation with Doctor and Hospital Recommendation System

With the advancement in technology, there is a need to provide access to advanced healthcare facilities. Since there is a limit to the capacity of healthcare facilities, minimising the need to visit a hospital and the load on healthcare facilities is essential. This is more so during the COVID-19 pandemic.

Final Project:

In this project, an end-to-end system of tele-imaging, video consultation with a doctor as well as hospital recommendation were implemented.

In the event of suspected COVID-19 cases, chest x-ray images and CT scans are automatically sent to an available doctor for video consultation.

If a hospital visit is necessary, recommendation of hospital will be made, based on availability of beds.
 

Student: 
Mr Khare Shubham Hari
Specialist Diploma in Applied Artificial Intelligence

Study of Customer Churn using Data Analytics and Machine Learning

One of the top UK telecom providers noticed a growing trend in customers leaving for the competitors. Customer churn is always a major problem in the telecom industry as the cost of recruiting a new customer is more expensive than maintaining an existing client.

A high churn rate not only reduces the profit of the company, but it also indirectly increases the revenue of its competitors. 

Proposed Analytical Model: 

To reduce customer churn, the telecom has commissioned a study to explore six years of past data for meaningful insights, and to develop an analytical model that can: 

• Identify factors that make a customer churn; and 

• Identify customers who are at higher risk of churning.

Using CRISP Data Mining Methodology, business understanding, exploratory data analysis and data cleansing were carried out.

Various types of decision trees, regression and neural network models were also developed. Finally, the regression model, with the highest precision, is chosen as the model for the deployment and prediction of churn.
 

Student: 
Ms Jenny Octavia Tjuwita
Specialist Diploma in Business Analytics

Back to top of page

2019

Analysis of Service Quality & Guest Experience in Popular Singapore Hotels

13,000 survey customers’ feedback (structured and unstructured text) were collected from 2005 – 2014 from 5 hotels in Singapore, namely, MBS, Fullerton, Raffles, Holiday Inn and IBIS.

Main Aim of the Project: 

The objective of the project is to answer the following business questions: 

• What are the differences in the profiles of reviewers who have stayed in the five different hotels throughout the surveyed years? 

• Which are the top-ranking hotels based on the hotel ratings? 

• Which are the factors that affect (or even predict) the rating of hotels? 

• What are the major concerns that reviewers tend to highlight during their stay in the different hotels? 

• What is the overall sentiment of reviewers when staying in a Singapore hotel?

Modeling and Visualisation Solution:

Descriptive Statistics and Data Visualisation are applied to explore the raw data before the appropriate data transformation is carried out. Transformed structured data are analysed using Statistical Predictive Modelling techniques.

For unstructured data like textual feedback, Text Mining techniques are applied to discover the key themes and sentiments of the feedback 

Technologies:

• Descriptive Statistics 

• Statistical Models (Regression and Decision Trees) 

• Text Mining, Text Clustering 

• Text Topic Extraction 

• Text Sentiment Analysis
  
  

Student:
Ravindran Amirthalinga
Specialist Diploma in Business Analytics

Credit Card Fraud Detection

With legacy and current enterprise systems, financial institutions are unable to precisely identify or classify fraud credit card transactions. In order for them to meet business need, they are in need of improving their system to identify or classify the fraud transactions precisely. 

Requirements: 

To resolve this problem, organisations need to leverage AI - Machine Learning Technology, which helps them to analyse and learn from the vast amount of data they have gathered so far. 

AI Based Solution: 

With the data gathered, they can build good dataset. Machine Learning Technology provides a lot of algorithms, which can be used to learn from the dataset and generalise unseen data precisely. 

Technologies:

Python 3.7, NumPy, Pandas and Scikit Learn Framework
 

Student:
Kandasamy Subbaian Karthik
Specialist Diploma in Applied Artificial Intelligence

Loan Prediction

Loans are one of the more common financial products offered by banks & finance houses. They are always trying to figure out the most effective business strategies to persuade customers to apply for their loans. In spite of the checks and balances put in place for loans, there are some customers that prove to be bad investments after their applications have been approved.

To reduce the chances of too many borrowers defaulting, financial institutions have to find some method to predict customers’ behaviours. Machine learning algorithms have proven to have good performance. 

Requirements: 

• A front-end web form to get user data
• A back-end model used for prediction (loan approved or denied) 

Proposed Solution: 

• Use historical data to build the model
• Experiment with the different models
• Evaluate the different models to find the best performing one which will be deployed in production

Technologies:

Numpy, scikit-learn, Pandas, Seaborn, Matplotlib, Jupyter Notebook, Anaconda 3, Flask, Pickle, XGBoost, Logistic Regression, Voting Ensembles, Boosting, Bagging, Decision Trees, Logistic Regression, Random Forest, Web Services
 

Student:
Adelene Ng
Specialist Diploma in Applied Artificial Intelligence

Soul Food Bot

Choosing a restaurant from the mountain of listings and user reviews on business directory websites such as Yelp can be a daunting if not intractable effort. How can one pour through thousands of reviews and ratings and make sense of them all to reach a decision? 

Requirements:

Design a recommender chatbot that can perform natural language processing so that a user only need to tell the chatbot what he desires or dislike for a restaurant experience, and the chatbot will make the recommendations. 

Final Solution:

Design and develop a chatbot with a recommendation system that takes in a composite of existing ratings, fresh input from user sentiments, likes and dislikes, together with a predicted user’s rating from Random Forest Classifier model to produce a hybrid recommendation system that can generate a list of recommendations to the user. 

Technologies:

Machine Learning, Natural Language Processing, Recommendation System, Dialogflow, Telegram
 

Student:
Liang Chee Wei, Kenneth
Specialist Diploma in Applied Artificial Intelligence

Applying Unsupervised Machine Learning to Detect Anomalous Network Traffic for Cyber Security Monitoring

Conventional cyber security monitoring involves crafting detection rules in SIEM to match specific patterns within the logs of interest. Bulk of the ingested data in the SIEM are not responsible for any detections but archived for future investigation needs.

Furthermore, specific detection patterns are based on potential scenarios of known malicious activities gathered from best practices or framework on adversary techniques. Detection may not work against new techniques. 

Requirements: 

A machine learning solution to derive additional value from the ingested web proxy logs by generating new insights from the log data. Instead of signature or pattern-based detection, the solution detect anomalies or outliers within the web proxy logs by leveraging on machine’s ability to learn about the data.

The flagged anomalies or outliers will be investigated by security analysts. Anomaly score would be assigned to the results to allow security analysts to prioritise their investigations. 

Proposed Solution: 

Apply unsupervised machine learning, Isolation Forest and Autoencoder, for Anomaly Detection On Web Proxy Logs. 

Technologies:

Python H2O framework, Tensorflow, Universal Sentence Encoder
 

Student:
Rick Tan
Specialist Diploma in Applied Artificial Intelligence

Product Price Suggestion for Sellers on eCommerce Platforms

Sellers in eCommerce platforms often face the problem of needing to spend time to research on what prices are set by competitors selling the same or similar products, and gauging how much they need to adjust their prices to remain competitive.

Due to information asymmetry, sellers often do not price their items well if they do not do enough research, and this creates unnecessary disparity in prices for similar products - those who are not pricing competitively usually lose out in sales revenue, which is not beneficial to both seller and eCommerce platforms. 

Main Aim of the Project: 

• This project aims to provide eCommerce sellers with suggested prices for their new or existing listings on eCommerce platforms. Ideally, it should be integrated with eCommerce platforms for a seamless experience for sellers when they are creating or managing their listings.

• The project is currently in the form of a Minimum Viable Product (MVP), with more focus on working product than aesthetics or user experience for the initial phase. 

Final Solution: 

• A web application for sellers to enter item information and obtain suggested prices for the item. The price suggestion is based on historical prices of items on the platform similar to the item the user has input to the form.

• The value proposition to sellers is the ability to get a quick view of whether their prices are competitive for existing listings and suggest a price for their new items that customers would be likely to buy. 

Technologies:

lask, keras, NLTK, pandas, sklearn, graphviz, matplotlib 
 

Student:
Quek Chiew Xia
Specialist Diploma in Applied Artificial Intelligence

YelperAssistant

Yelp is a search service using crowd-sourced reviews about local businesses and facilitates searching for events, lists and communication between Yelp users.

Problem Statement:

Non-user-friendly search interface coupled with generic information not personalised to user’s interest resulting in onerous search experience.

Key Requirements:

• To develop a personalised recommender based on reviews data 

• To provide a virtual assistant interface to allow users to query for personalised recommendations and to facilitate Yelp information search

Solution:

Slack/Telegram chatbot powered by Google DialogFlow with fulfilment by Yelp Fusion API (for Yelp information) and recommendation engine powered by Scikit-Surprise. 

Technologies:

• Google

• DialogFlow 

• Scikit-Surprise 

• Yelp Fusion API 

• Flask 

• REST API 

• micro-services
   

Student:
Lim Yuan Her
Specialist Diploma in Applied Artificial Intelligence

IKEA@Alexandra Shuttle Bus Virtual Assistant

Shoppers commented that it is inconvenient to get to IKEA@Alexandra from the nearest MRT stations, as the interval of public bus service available along this route outside IKEA@Alexandra, like bus service 195, is too long and inaccurate. The buses also get too crowded, especially during weekends. Additionally, shoppers are not aware of in-store promotions and sales. 

Requirements: 

• To assist potential shoppers in reaching IKEA@Alexandra more efficiently, the free shuttle bus service operates during weekends, providing a more accurate Estimated Time of Arrival (ETA). Additionally, on weekdays, SBS Transit bus service number 195 operates from Queenstown and Tiong Bahru MRT Stations, further enhancing the value-added services provided through virtual assistant technology.

• Another added business value is to inform potential and existing shoppers about current promotions, sales and deals available in-store.

Proposed Solution: 

• Deploy a GPS tracker onboard IKEA's free shuttle bus services to share its location with the Google Maps Direction API, allowing it to compute the Estimated Time of Arrival (ETA) to designated MRT pickup points.

• Utilise LTA DataMall API to provide ETA of public bus service 195 at designated MRT pickup points.

• Use of chatbot to disseminate promotions, sales and deals. 

Technologies:

Google’s Dialogflow, with integration to Telegram chatbot, Google Maps Direction API, LTA DataMall API, MQTT Broker with publication and subscription services, Single Board Computer (SBC) with GPS module
 

Student:
Wong Chun Yun
Specialist Diploma in Applied Artificial Intelligence

Personalising Consumer Experience for Marriott Hotels

In today’s age of dynamic digital disruption, consumers are becoming not just more connected but hyper connected. The hospitality industry is one of the most dynamic, brutal, unyielding competitive sectors where consumers are at the heart of its operations.

For a very long time, there was almost no innovation in this field, but the arrival of technology-fueled on-demand economy has dramatically changed that. Even traditional restaurant chains began experimenting, growing, acquiring and innovating intensely. 

Requirements: 

Hospitality sectors remain a key for economic development and job creation around the world. There is a strong need to increase food and beverage sales in restaurants, bars and fast-food outlets within hotels.

In the era of the hyper-connected consumers, delivering a personalised experience is key. Consumers are connected to multiple devices that range from desktop, laptop, mobile, smart devices, wearables, car navigations, entertainments and even home appliances. All this generates humongous information about individuals and can be utilised to know your consumers better.

Today, most of the information technology companies are investing in Artificial Intelligence (AI) and Machine Learning (ML) to produce meaningful information out of big data and provide the relevant recommendations. 

Final Solution: 

• This project provides the solution to recommend food items and services using collaborative and content-based filtering techniques.

• Recommendations are more personalised based on the users and item ratings. 

Tools & Technologies:

Python 3.7 and Python libraries such as numpy, pandas, scikit-learn, seaborn, matplotlib: Jupyter Notebook, Dialogflow, Pythonanywhere.com
 

Student:
Vartak Sachin Avinash
Specialist Diploma in Applied Artificial Intelligence

Building a Hotel Recommendation System using Collaborative Filtering with Benchmarking of Algorithms

Main Aim of the Project: 

This project aims to build a hotel recommender system using Tripadvisor dataset that aims to recommend top ten hotels based on collaborative filtering method with two different approaches. 

Requirements:

• The project is developed using machine learning libraries from Surprise package and scikit-learn packages.

• The algorithms used are based on collaborative filtering method of recommendation systems such as baseline algorithms, KNN based and Matrix factorization-based algorithms. 

• RMSE and MAE scores were used as the accuracy metric for the predictions. 

Proposed Solution: 

The recommender system was built and tested using two approaches:

1. Using user ratings

2. Using both review test and user ratings

Both approaches are able to generate top 10 recommended hotels based on the user’s needs. 

Technologies:  

• Python 

• Jupyter notebook 

• Python machine learning libraries from surprise

• Scikit learn packages 
   

Student:
Sekar Karthik
Specialist Diploma in Applied Artificial Intelligence

Back to top of page