Machine Learning series – Part II
The Machine Learning Series Part 2 had two presentations. The first was by Christo Strydom – MyBucks. He presented on fraud detection in an online environment. This talk revolved around a journey of one company in their attempt to assess the extent of fraud and identifying fraud in real time. The talk showed case studies of identity theft and illustrated the use of Neo4J as a graphing database and document store to achieve detection goals. Clickstream analytics were used in an illustration with the aim of arriving at a (fraud) model to score applicants on the likelihood of being fraudulent. Classification schemes of client behaviour and logic were discussed and early attempts at a deep learning of the problem were shown.
The second presentation was by Robert Berman (CSIR) on Hyperformance: predicting high-speed performance of a B-Double. This talk gave an overview of a data driven approach to develop a prediction model for the high-speed dynamic performance of common heavy goods vehicle (HGV). Robert gave an overview of the current performance-based standards (PBS) approach to HGV design and regulation, followed by how machine learning could assist and inform this process. The talk further covered the data collection, post-processing, and model fitting. Robert discussed the use of Bayesian optimisation for hyperparameter optimisation and how this data science approach to HGV design benefits both industry and regulators.
Machine Learning series – Part I
Two further seminars, Machine Learning series – Part I and Part II were held together with the Operations Research Society of South Africa (ORSSA) on 11 May 2017 and 1 August 2017 respectively at Resolve solutions Partners, Samrand.
On 11 May, Marc Gagiano presented a talk on Churn prediction. He gave a brief introduction of data science in the context of business highlighting how machine learning fits in. There was a brief discussion of the differences between machine learning and regression. A case study for a telecommunication company was presented where machine learning was used to generate more revenue. The main application was churn prediction using neural networks and confusion matrices were used to measure the performance in order to quantify the additional revenue generated. The algorithm and variations thereof were explained along with details on the implementation thereof.
A second presentation on 11 May was given by Christiaan Hattingh (GetBucks) on Real time credit decision as a machine learning classification problem. This case study was on granting micro-credit. Christiaan discussed how credit risk decisions can be viewed as a machine learning classification problem, with a brief, non-technical description of machine learning algorithms and approaches for this purpose. The real-time implementation and challenges were discussed, as well as how machine learning allows quantitative adjustment of risk appetite, with a calculated expectation of returns.
ICCSSA breakfast seminar – 28 October 2016
A breakfast seminar was held on 28 October 2016 at SAS institute, jointly hosted by ICCSSA, ORSSA and SAS. The first talk was given by Dr Vincent Micalli, currently the CEO of Stats4Buz (Pty) Ltd, on “A long walk from facts and figures to predictions in the electrical energy industry”.
The second talk was given by Dr Zaid Kimmie, currently responsible for Planning and Strategy at CSIR, on “Elections forecasts: the CSIR and the 2016 Local Government Elections”. The seminar was sponsored by SAS Institute.
ICCSSA breakfast seminar – High RE Scenario – 2 September 2016
On 2 September 2016 ICCSSA held a joint breakfast seminar with ORSSA, UNISA and SASA. The first talk was given by Dr Tobias Bischof-Niemz, Centre Manager for Energy at CSIR, on “New high-fidelity wind and solar PV data for South Africa enable renewables-based planning”
The second talk was given by Dr Caston Sigauke, senior lecturer at the Statistics Department at the University of Venda on “Short term hourly load forecasting during the peak period using quantile regression with an application to the unit commitment problem”. The seminar was sponsored by ICCSSA and ORSSA and the venue was sponsored by UNISA. The mix between industry and academia created a lively and vibrant discussion.
ICCSSA second breakfast seminar – 4 December 2015
ICCSSA launched its second breakfast seminar on 4 December 2015 in partnership with the Centre of Excellence for Mathematical and Statistical Sciences (CoE-MaSS) at Wits University, Operations Research Society of South Africa (ORSSA), South African Statistical Association (SASA) and the University of Pretoria (UP). The guest speaker was Prof Alfred Stein from the Netherlands. He spoke on “Spatial statistical aspects of big data: new challenges for research”.
ICCSSA first breakfast seminar – 24 July 2015
On 24 July 2015 ICCSSA held its first breakfast seminar. The first talk was given by Professor Paul Fatti on “Big Data, Data Science and Analytics – The End of Statistics”. It was well received and initiated a very lively discussion.
The second talk was given by Mr Ben Mphahlele the Chairperson of the South African Statistics Council on the “Need for Change”. The event was sponsored by SAS and SACNASP. The ICCSSA Board decided that over the next few years the ICCSSA convention would be replaced by regular ICCSSA seminars.
First ICCSSA convention held in Milnerton Cape Town, 28 – 30 March 2012
The first ICCSSA conference was held at the Lagoon Beach Hotel in Milnerton, Cape Town from 28 – 30 March 2012. It was attended by 50 delegates from a variety of fields of specialisation…