Team5_1A_Fradu

Inspiration

We noticed a lot of fraud cases, including our loved ones and friends, and we want to prevent the escalation of scam cases. Banks are not very efficient in dealing with these scam cases after they had happened, and many people's hard-earned money are cheated away from them. To tackle this, we believe prevention is better than cure and developed Fradu.

What it does

It is a telegram bot which implements Trainable Artificial Intelligence to determine the credibility of text messages from a scale of 0 to 1. Users are able to forward any messages to the bot. Furthermore, greatly help the digitally less-literate persons like our nannies.

How we built it

We first learnt how to build a telegram bot from Youtube as we wanted to incorporate AI agents to our project. We then wanted to have an AI system to detect fraud messages. We made use of the binary classification model, notably the Multinomial Naive Bayes classifier, to meticulously assess the credibility of messages. The classifier, based on the principles of Naive Bayes, learned patterns and features indicative of potential scams. We utilized a dataset containing labeled examples of legitimate and suspicious messages to train the model. The Telegram chatbot interface was integrated with the trained classifier, enabling real-time analysis of forwarded messages. The algorithm's predictions on message credibility assist users in identifying potential scams. This technical synergy of machine learning and chatbot technologies forms the core of "Fradu's" functionality, providing a robust defense against financial fraud.

Challenges we ran into

It was our first time training an Artificial Intelligence bot and we used too little binary classification cases which resulted in many normal messages getting flagged as suspicious and many suspicious messages passing through the bot as legitimate. We included many train cases for each side, for example, investment scam, part-time job scam, financial info scam messages to better train our bot to detect suspicious messages and also included many normal messages as well.

We did not know how to integrate a telegram bot with an Artificial Intelligence framework and had to research extensively and use ChatGPT to our fullest and eventually we made it work.

Accomplishments that we're proud of

We were able to code a telegram bot and AI from zero experience and knowledge, along with the video introduction in this very short time frame. We were able to integrate these two things together despite our initial lack of experience.

What we learned

We learnt the basics of making a Telegram bot and AI framework from scratch and learnt and made use of pandas as well to use online datasets for our program. However, we couldn't find a good online dataset, so we end up not using Pandas. We learnt about the Sklearn extension of python as well.

What's next for Fradu

Fradu will get more training and be able to differentiate messages with very high accuracy. Besides, Fradu can collaborate with security experts in telegram for further improvements like reporting fraud accounts automatically as scammers tend to use this platform the most due to its feature of allowing people to remain anonymous without leaking any contact information unlike WhatsApp or Facebook. It can continue to scale to be an app to detect fraudulent messages from all messaging apps and finally, result in a decrease in money lost through scams.