Niral J Shah

An ELO-rating model for loan default prediction

2020-05-21T00:00:00Z

ELO-rating is used to predict the performance of a sport player in a tournament or otherwise. This works as defining an ideal or an average model player and comparing the performance measures of all other players with this model player. This gives us an overview of how each player is better than each other indirectly.

This can be also work in finance. Although, no company is reportedly using this kind of technique, I believe we can create an ideal loan applicant by taking the average of all the loan applicants who have successfully repayed their loan. This will give us an idea of what an average non-defaulting loan applicant looks like and we can compare all the loan applicants with this model applicant.

This will not only reduce the need of credit report. This is a huge change in the cost per customer.

The best part about the process is that it will give out an ELO-rating or a new credit score along with a credit report with a CONFIDENCE INTERVAL. This means that when we are not very confident on our outcome (which is less than 7% of the time), we can work our way through by getting an actual credit report for the applicant. The potential of this task when tested on the demo data showed us that it can decrease the rate of defaulting from 8.5% to 1.8%

Stock Price Analyzer using sentiment analysis on news articles and tweets

2019-08-22T00:00:00Z

News and current affairs affect the stocks when it comes to trading. It also affects the way people think & the way they make a decision. We have a seen that the sentiment behind the news affect more than it should be. But how can we escape it and predict the change before it happens?

Bloomberg is already using the technique of using sentiment analysis on current news to estimate the change on related stocks. And making millions out of it.

You can find this project and its code on Github here

Here is an example of how it works...

I used the Business Times to search for news articles on Facebook. Ideally, we can use tweets from powerful celebrities, media houses and companies to reach to the news faster. So I ran it for the last six months and averaged the sentiments of each news article for a given day. Here it goes...

Date	Sentiment rate
26 Jul 2019	0.796
25 Jul 2019	0.981
24 Jul 2019	−0.086
18 Jul 2019	0.883
17 Jul 2019	0.962
16 Jul 2019	0.986
15 Jul 2019	−0.548
13 Jul 2019	0.38
12 Jul 2019	−0.176
11 Jul 2019	0.684
10 Jul 2019	−0.428
09 Jul 2019	0.246
04 Jul 2019	−0.264
03 Jul 2019	0.094
02 Jul 2019	0.055
01 Jul 2019	−0.243
28 Jun 2019	−0.006
27 Jun 2019	0.842
26 Jun 2019	−0.194
25 Jun 2019	−0.27
21 Jun 2019	−0.357
20 Jun 2019	0.997
19 Jun 2019	0.814
18 Jun 2019	0.978

Here, every sentiment score for the day shows the sentiment average for the day. 1 means a very positive news for the company while -1 tends to show the negativity or a setback for the company. This gives an idea about how the stock of facebook is going to be in the near term. Let us see how effective it is:

Facebook’s share price, volume and daily news sentiment, June–July 2019.

In the above image we can see that the dip in the second week of July 2019 on Facebook stock could have predicted. The bull run for 26th June till 10th July 2019 could be predicted as the last point on Facebook news was positive on 25th June 2019 Thereafter, the sentiments are fairly stable near zero value which indicated there will be a stable change all the way till next non-zero sentiment value.

Further steps of improvement here would be to devise an relatability algorithm as sometimes the news article does not always affect the stock. The other main step would be to figure out the accuracy of this method and how to improve it.

Fall Relief: An effort to reduce falls and related injuries among age group 65 and older

2019-08-04T00:00:00Z

Most people consider falls are inevitable in seniors. Wrong! Contrary to the common belief, in most cases accidental falls are preventable.

Globally, falls are a major public health problem. For people aged 65 years or older, the average health system cost per fall injury in Australia is a substantial A$ 1630.

Implementation of effective prevention strategies with a subsequent 25% reduction in the incidence of falls for the 65+ age group could create a net savings of over A$ 50 million.

Our product "Fall Relief" is born to solve this issue and improve user's lifestyle, by providing them with a targeted lifestyle plan which includes exercises and dietary regimes catered based on the user's current deficiency to help them live a better life and reduce the likelihood of falling.

Made by a group of 4: 2 Business Analyst, an IT Developer and a Data Scientist - Team BigHD

The Fall Relief site, as the team built it in 2019.

Humanoid Robots

2017-12-22T00:00:00Z

A guest article on robots built to look and move like people — how they are put together, where artificial intelligence comes in, and where they are already useful, from healthcare to the home. (Circuit Digest)

What is a Hoverboard and How does it work?

2017-11-03T00:00:00Z

A guest article on how a self-balancing scooter stays upright — the accelerometers, gyroscopes and magnetometers inside it, the parts it is built from, and how it is steered. (Circuit Digest)