Hi Varun
I've been actively designing, creating and deploying AI Predictive Solutions using machine learning techniques including neural networks for over 25 years for telecommunications, banking, mineral exploration, quality control, speech recognition, ... and one main piece of advice I'd give would be that it is not so much the AI Predictive Analytics package itself but the data integration and transformation that is key to producing an effective solution delivering business benefit. One recent data scientist I talked with at a large telco company in Europe said they spent 80-90% of their time just getting the data into the required structure for modelling. This was especially true for large volume transaction data (e.g. CDRs, recharges/adjustments, etc.). Another important aspect for delivering a solution is to be able to produce results with enough supporting information to allow an end-user to trust the results being produced - years ago I remember regular meetings with clients where the "Black Box" issue would be raised; our company Neural Technologies provided solutions to this issue back then... it is interesting how Google et al are also now recognising this need with the concept of "Explainable AI". Many off-the-shelf predictive libraries lack support for this capability as they've mainly been produced with a focus on delivering predictive accuracy.
Our Optimus Platform has been produced with a strong focus on telecommunications and provides awesome flexibility with regards data integration and transformation to enable predictive analytics using machine learning. As you might imagine given our company name we equally provide tried and trusted machine learning algorithms... with predictive analytics given it is highly Data Driven it is also worth bearing in mind it is equally sensitive to data - again techniques to provide effective machine learning monitoring in Production is incredibly important. e.g. when predicting credit worthiness for telco customer application processing it would be embarrassing to say the least if poor predictive performance was only detected 6+ months down to the line after write-off processes were completed. Having methods to identify that a model is behaving in a non-stable fashion is vital. Hence the overall picture of pre-model data aquisition/transformation -> modelling -> live model monitoring as a 3-stage framework is key to delivering a trusted solution. We had a great example recently for one of our customers where a large roaming abuse was identified via machine learning techniques that normal rules-based analysis would either never, or only very late, have detected.
Feel free to contact me for more info - I'm not a sales guy either!
best regards
George Bolt
Head of Analytics, Neural Technologies
------------------------------
George Bolt
Neural Technologies
------------------------------
Original Message:
Sent: Apr 21, 2020 03:15
From: Varun Pandhi
Subject: Packages for AI Predictive Analytics
Hi all,
Can you please suggest some good packages catering to AI Predictive Analytics (both paid and free Open Source)?
They should be suited to Telecommunication needs and aligning to AI standards (if any laid out by TMF) as far as possible.
#AIandData
#TMForumGeneral
------------------------------
Varun Pandhi
------------------------------