Re: Machine Learning Requirements Discussion: WAS [Introducing Maha Refai, Mifos Initiative volunteer, working on requirements for credit analytics]

[Lalit Mohan S]

Dear All,

My 2 cents...

For training set, we need to take the help of experts, banks, regulators
and use available public datasets.  Also, there are some research papers on
Fraud detection and other models, we need to connect with some of the
academic researchers as well.

I had some recent conversation with a manger at RBI(India's central bank),
he expressed interest to share his experiences and patterns that he
observed on how people (banks and customers) circumvent the rule based NPA
calculation. Not to be quoted elsewhere :)

regards
Lalit

On Wed, Apr 11, 2018 at 12:48 AM, Maha El-Refai 
wrote:

> Hello Ed/James,
>
> Sorry for the late reply, I'm traveling most of the time in different time
> zones.
>
> I agree with James, I had a couple of calls to collect ideas of how can
> machine learning help us and my only concern was our access to the data and
> how structured and classified is the data. If there will be a need to
> annotate or map the data manually and what's the size of the data that we
> can extract?
>
> If anyone has an input here that would be very helpful.
>
> Thanks,
> Maha
>
>
> On Wed, Apr 4, 2018 at 11:00 PM, James Dailey 
> wrote:
>
>> Hi Maha -
>>
>> I've been spending some time thinking about this for some business areas
>> adjacent to financial inclusion.
>>
>> In financial inclusion, and financial services in general, there are of
>> course, many areas where some ML or even basic analysis would do a
>> wonderful job.
>>
>> For Mifos I believe a key thing that is needed is the ability to extract
>> data in meaningful structures (flattened in appropriate ways) and within a
>> reasonable taxonomy of credit, accounting, and consumer terms.  Perhaps
>> that means making data more easily discoverable via APIs.  I also suspect
>> that permission and rights management to data - with screens for privacy
>> (or anonymizing) are needed.
>>
>> Related, one concept I've come across is the ability for third parties to
>> get the "results" of machine learning models without actually getting
>> access to the data and to apply that in a kind of federated data approach -
>> where the models are trained up and then "call home" with the results.
>> This would imply that it is important for the data access APIs to be
>> structured consistently and with backwards compatibility.
>>
>> For ML to work there needs to be some other "big" data sets related
>> together.  You may be interested in a recent world bank conference where a
>> bunch of new ML applications were introduced:  https://blogs.wor
>> ldbank.org/developmenttalk/artificial-intelligence-economic-
>> development-conference-roundup-27-presentations
>>
>> Happy to get on a call at some point.
>>
>> Thanks,
>> James
>>
>>
>> On Wed, Apr 4, 2018 at 10:45 AM Ed Cable  wrote:
>>
>>> Maha,
>>>
>>> I wanted to resurrect this thread and invigorate some discussion as we
>>> have
>>> a GSOC prospect, Lalit, considering a project focused around machine
>>> learning. I've added him to this thread and going to also add the team at
>>> BowPI so they can contribute to the discussion around what they've done
>>> on
>>> the machine learning front to date.
>>>
>>> Ed
>>>
>>> On Thu, Nov 9, 2017 at 4:58 PM, Avik Ganguly 
>>> wrote:
>>>
>>> > Hi Maha,
>>> >
>>> > Welcome to the community! Cognitev looks like an exciting startup. The
>>> > current developments going on in Fineract; i.e. the upcoming Fineract
>>> CN
>>> > application probably makes this the right time to initiate this effort
>>> as
>>> > ML services usually needs to scale separately and the independence of
>>> > choosing a separate tech stack for these services matters a lot.
>>> >
>>> > Adding a bit on one part that Ed wrote - "monitor ongoing credit
>>> risk". In
>>> > the microfinance space, the most recurring problem statement I have
>>> heard
>>> > from risk heads in India is collusion by loan officers to meet targets
>>> with
>>> > / without ringleaders which is addressable through early warning
>>> systems.
>>> > Looking forward to working with you.
>>> >
>>> > With best regards,
>>> > Avik Ganguly.
>>> >
>>> > On Wed, Nov 8, 2017 at 9:02 PM, Ed Cable  wrote:
>>> >
>>> > > Hello community,
>>> > >
>>> > >
>>> > > I wanted to introduce to you a new Mifos Initiative volunteer, Maha
>>> Refai
>>> > > , who will be reaching out
>>> to
>>> > > members of the community to document their requirements around
>>> credit and
>>> > > risk analytics using artificial intelligence, machine learning and
>>> big
>>> > > data.
>>> > >
>>> > > Maha Refai is currently the co-founder and Head of Product at
>>> Cognitev, a
>>> > > startup based in UAE and Egypt using artificial intelligence to
>>> automate
>>> > > digital marketing. After receiving her bachelor’s in Computer
>>> Science she
>>> > > spent the past ten years primarily in product and project management
>>> in
>>> > the
>>> > > entertainment industry while also getting her MBA.
>>> > >
>>> > >
>>> > > She wi

Re: Machine Learning Requirements Discussion: WAS [Introducing Maha Refai, Mifos Initiative volunteer, working on requirements for credit analytics]

[Maha El-Refai]

Hello Ed/James,

Sorry for the late reply, I'm traveling most of the time in different time
zones.

I agree with James, I had a couple of calls to collect ideas of how can
machine learning help us and my only concern was our access to the data and
how structured and classified is the data. If there will be a need to
annotate or map the data manually and what's the size of the data that we
can extract?

If anyone has an input here that would be very helpful.

Thanks,
Maha


On Wed, Apr 4, 2018 at 11:00 PM, James Dailey 
wrote:

> Hi Maha -
>
> I've been spending some time thinking about this for some business areas
> adjacent to financial inclusion.
>
> In financial inclusion, and financial services in general, there are of
> course, many areas where some ML or even basic analysis would do a
> wonderful job.
>
> For Mifos I believe a key thing that is needed is the ability to extract
> data in meaningful structures (flattened in appropriate ways) and within a
> reasonable taxonomy of credit, accounting, and consumer terms.  Perhaps
> that means making data more easily discoverable via APIs.  I also suspect
> that permission and rights management to data - with screens for privacy
> (or anonymizing) are needed.
>
> Related, one concept I've come across is the ability for third parties to
> get the "results" of machine learning models without actually getting
> access to the data and to apply that in a kind of federated data approach -
> where the models are trained up and then "call home" with the results.
> This would imply that it is important for the data access APIs to be
> structured consistently and with backwards compatibility.
>
> For ML to work there needs to be some other "big" data sets related
> together.  You may be interested in a recent world bank conference where a
> bunch of new ML applications were introduced:  https://blogs.
> worldbank.org/developmenttalk/artificial-intelligence-
> economic-development-conference-roundup-27-presentations
>
> Happy to get on a call at some point.
>
> Thanks,
> James
>
>
> On Wed, Apr 4, 2018 at 10:45 AM Ed Cable  wrote:
>
>> Maha,
>>
>> I wanted to resurrect this thread and invigorate some discussion as we
>> have
>> a GSOC prospect, Lalit, considering a project focused around machine
>> learning. I've added him to this thread and going to also add the team at
>> BowPI so they can contribute to the discussion around what they've done on
>> the machine learning front to date.
>>
>> Ed
>>
>> On Thu, Nov 9, 2017 at 4:58 PM, Avik Ganguly 
>> wrote:
>>
>> > Hi Maha,
>> >
>> > Welcome to the community! Cognitev looks like an exciting startup. The
>> > current developments going on in Fineract; i.e. the upcoming Fineract CN
>> > application probably makes this the right time to initiate this effort
>> as
>> > ML services usually needs to scale separately and the independence of
>> > choosing a separate tech stack for these services matters a lot.
>> >
>> > Adding a bit on one part that Ed wrote - "monitor ongoing credit risk".
>> In
>> > the microfinance space, the most recurring problem statement I have
>> heard
>> > from risk heads in India is collusion by loan officers to meet targets
>> with
>> > / without ringleaders which is addressable through early warning
>> systems.
>> > Looking forward to working with you.
>> >
>> > With best regards,
>> > Avik Ganguly.
>> >
>> > On Wed, Nov 8, 2017 at 9:02 PM, Ed Cable  wrote:
>> >
>> > > Hello community,
>> > >
>> > >
>> > > I wanted to introduce to you a new Mifos Initiative volunteer, Maha
>> Refai
>> > > , who will be reaching out to
>> > > members of the community to document their requirements around credit
>> and
>> > > risk analytics using artificial intelligence, machine learning and big
>> > > data.
>> > >
>> > > Maha Refai is currently the co-founder and Head of Product at
>> Cognitev, a
>> > > startup based in UAE and Egypt using artificial intelligence to
>> automate
>> > > digital marketing. After receiving her bachelor’s in Computer Science
>> she
>> > > spent the past ten years primarily in product and project management
>> in
>> > the
>> > > entertainment industry while also getting her MBA.
>> > >
>> > >
>> > > She will be reaching out to some of you directly and documenting her
>> > > findings on the wiki. But if you'd like to be part of this
>> requirements
>> > > gathering process, please reply to this thread and we'll set up a
>> call.
>> > >
>> > > Based on her knowledge and expertise of artificial intelligence and
>> > machine
>> > > learning, she is going to help  develop a product management strategy
>> for
>> > > both the current Apache Fineract platform and the upcoming Apache
>> > Fineract
>> > > CN application framework that identifies and prioritizes what are the
>> > > necessary microservices and modules that are part of the framework or
>> > > integrated with it to facilitate the capture of data, the analysis of
>> it,
>> > > and the ability to m

Re: Machine Learning Requirements Discussion: WAS [Introducing Maha Refai, Mifos Initiative volunteer, working on requirements for credit analytics]

[James Dailey]

Hi Maha -

I've been spending some time thinking about this for some business areas
adjacent to financial inclusion.

In financial inclusion, and financial services in general, there are of
course, many areas where some ML or even basic analysis would do a
wonderful job.

For Mifos I believe a key thing that is needed is the ability to extract
data in meaningful structures (flattened in appropriate ways) and within a
reasonable taxonomy of credit, accounting, and consumer terms.  Perhaps
that means making data more easily discoverable via APIs.  I also suspect
that permission and rights management to data - with screens for privacy
(or anonymizing) are needed.

Related, one concept I've come across is the ability for third parties to
get the "results" of machine learning models without actually getting
access to the data and to apply that in a kind of federated data approach -
where the models are trained up and then "call home" with the results.
This would imply that it is important for the data access APIs to be
structured consistently and with backwards compatibility.

For ML to work there needs to be some other "big" data sets related
together.  You may be interested in a recent world bank conference where a
bunch of new ML applications were introduced:
https://blogs.worldbank.org/developmenttalk/artificial-intelligence-economic-development-conference-roundup-27-presentations


Happy to get on a call at some point.

Thanks,
James


On Wed, Apr 4, 2018 at 10:45 AM Ed Cable  wrote:

> Maha,
>
> I wanted to resurrect this thread and invigorate some discussion as we have
> a GSOC prospect, Lalit, considering a project focused around machine
> learning. I've added him to this thread and going to also add the team at
> BowPI so they can contribute to the discussion around what they've done on
> the machine learning front to date.
>
> Ed
>
> On Thu, Nov 9, 2017 at 4:58 PM, Avik Ganguly 
> wrote:
>
> > Hi Maha,
> >
> > Welcome to the community! Cognitev looks like an exciting startup. The
> > current developments going on in Fineract; i.e. the upcoming Fineract CN
> > application probably makes this the right time to initiate this effort as
> > ML services usually needs to scale separately and the independence of
> > choosing a separate tech stack for these services matters a lot.
> >
> > Adding a bit on one part that Ed wrote - "monitor ongoing credit risk".
> In
> > the microfinance space, the most recurring problem statement I have heard
> > from risk heads in India is collusion by loan officers to meet targets
> with
> > / without ringleaders which is addressable through early warning systems.
> > Looking forward to working with you.
> >
> > With best regards,
> > Avik Ganguly.
> >
> > On Wed, Nov 8, 2017 at 9:02 PM, Ed Cable  wrote:
> >
> > > Hello community,
> > >
> > >
> > > I wanted to introduce to you a new Mifos Initiative volunteer, Maha
> Refai
> > > , who will be reaching out to
> > > members of the community to document their requirements around credit
> and
> > > risk analytics using artificial intelligence, machine learning and big
> > > data.
> > >
> > > Maha Refai is currently the co-founder and Head of Product at
> Cognitev, a
> > > startup based in UAE and Egypt using artificial intelligence to
> automate
> > > digital marketing. After receiving her bachelor’s in Computer Science
> she
> > > spent the past ten years primarily in product and project management in
> > the
> > > entertainment industry while also getting her MBA.
> > >
> > >
> > > She will be reaching out to some of you directly and documenting her
> > > findings on the wiki. But if you'd like to be part of this requirements
> > > gathering process, please reply to this thread and we'll set up a call.
> > >
> > > Based on her knowledge and expertise of artificial intelligence and
> > machine
> > > learning, she is going to help  develop a product management strategy
> for
> > > both the current Apache Fineract platform and the upcoming Apache
> > Fineract
> > > CN application framework that identifies and prioritizes what are the
> > > necessary microservices and modules that are part of the framework or
> > > integrated with it to facilitate the capture of data, the analysis of
> it,
> > > and the ability to make decisions upon the data.
> > >
> > > In short, volunteer will help identify how these predictive analytical
> > > tools along with big data can help make Mifos and Apache Fineract more
> > of a
> > > loan origination system, as well as means of developing a credit score
> to
> > > manage and monitor ongoing credit risk. You can see below for some more
> > of
> > > the specific tasks she’ll be undertaking.
> > >
> > >
> > >-
> > >
> > >Research on existing open source tools or libraries that could be
> > >integrated for machine learning such as weka.
> > >-
> > >
> > >Research on alternative credit scoring tools that are in existence
> > >-
> > >
> > >

Machine Learning Requirements Discussion: WAS [Introducing Maha Refai, Mifos Initiative volunteer, working on requirements for credit analytics]

[Ed Cable]

Maha,

I wanted to resurrect this thread and invigorate some discussion as we have
a GSOC prospect, Lalit, considering a project focused around machine
learning. I've added him to this thread and going to also add the team at
BowPI so they can contribute to the discussion around what they've done on
the machine learning front to date.

Ed

On Thu, Nov 9, 2017 at 4:58 PM, Avik Ganguly 
wrote:

> Hi Maha,
>
> Welcome to the community! Cognitev looks like an exciting startup. The
> current developments going on in Fineract; i.e. the upcoming Fineract CN
> application probably makes this the right time to initiate this effort as
> ML services usually needs to scale separately and the independence of
> choosing a separate tech stack for these services matters a lot.
>
> Adding a bit on one part that Ed wrote - "monitor ongoing credit risk". In
> the microfinance space, the most recurring problem statement I have heard
> from risk heads in India is collusion by loan officers to meet targets with
> / without ringleaders which is addressable through early warning systems.
> Looking forward to working with you.
>
> With best regards,
> Avik Ganguly.
>
> On Wed, Nov 8, 2017 at 9:02 PM, Ed Cable  wrote:
>
> > Hello community,
> >
> >
> > I wanted to introduce to you a new Mifos Initiative volunteer, Maha Refai
> > , who will be reaching out to
> > members of the community to document their requirements around credit and
> > risk analytics using artificial intelligence, machine learning and big
> > data.
> >
> > Maha Refai is currently the co-founder and Head of Product at Cognitev, a
> > startup based in UAE and Egypt using artificial intelligence to automate
> > digital marketing. After receiving her bachelor’s in Computer Science she
> > spent the past ten years primarily in product and project management in
> the
> > entertainment industry while also getting her MBA.
> >
> >
> > She will be reaching out to some of you directly and documenting her
> > findings on the wiki. But if you'd like to be part of this requirements
> > gathering process, please reply to this thread and we'll set up a call.
> >
> > Based on her knowledge and expertise of artificial intelligence and
> machine
> > learning, she is going to help  develop a product management strategy for
> > both the current Apache Fineract platform and the upcoming Apache
> Fineract
> > CN application framework that identifies and prioritizes what are the
> > necessary microservices and modules that are part of the framework or
> > integrated with it to facilitate the capture of data, the analysis of it,
> > and the ability to make decisions upon the data.
> >
> > In short, volunteer will help identify how these predictive analytical
> > tools along with big data can help make Mifos and Apache Fineract more
> of a
> > loan origination system, as well as means of developing a credit score to
> > manage and monitor ongoing credit risk. You can see below for some more
> of
> > the specific tasks she’ll be undertaking.
> >
> >
> >-
> >
> >Research on existing open source tools or libraries that could be
> >integrated for machine learning such as weka.
> >-
> >
> >Research on alternative credit scoring tools that are in existence
> >-
> >
> >Interviews with existing community members to discuss systems they
> have
> >implemented or how they’ve been able to do these data analysis using
> the
> >existing Mifos toolset.
> >-
> >
> >Interviews with existing community members who have desires to
> implement
> >credit scoring capabilities and big data analytics including
> stewarding
> >these conversations over our community mailing lists and documenting
> >requirements publicly
> >-
> >
> >Recommendations on a strategy of how to implement such features -
> either
> >as core microservices or ancillary modules.
> >-
> >
> >User Stories on actual features to implement within the framework or
> >product.
> >-
> >
> >Brainstorm and list out the insight and information that could be
> >analyzed so we can understand what data about the customer, their
> > behavior
> >and their loans must be captured in the framework and platform.
> >
> >
> >
> > --
> > *Ed Cable*
> > President/CEO, Mifos Initiative
> > edca...@mifos.org | Skype: edcable | Mobile: +1.484.477.8649
> > <(484)%20477-8649>
> >
> > *Collectively Creating a World of 3 Billion Maries | *http://mifos.org
> >   
> >
>



-- 
*Ed Cable*
President/CEO, Mifos Initiative
edca...@mifos.org | Skype: edcable | Mobile: +1.484.477.8649

*Collectively Creating a World of 3 Billion Maries | *http://mifos.org