Overview of select projects

Online Casinos

Modelling player behaviour

We predict players that are going to leave the casino before they actually leave. With this information, the marketing team can intervene with a retention strategy before their players leave. We also have the ability to predict how much different groups of players would spend in the future to assist (using our machine learning algorithms). Having this information on a client helps the casino focus there energy on the cluster of players that have the highest potential value. Our algorithms predict at an accuracy level greater than 80%.

Fraud Detection

Detecting personal and company-wide fraud

Financial statement fraud: Predicting companies that will receive qualified financial statements (possible financial statement fraud) is crucial to for the receiver of revenue, creditors and investors. Predictive analytics allows us to flag which companies have a greater probability of manipulating their financial statements in order to obtain an extended line of credit, inflate their share price and decrease their taxable income. Isazi Consulting has been able to predict financial statement fraud with an accuracy of above 80%.

Insurance fraud: Insurance companies need to determine the risk profile of individuals who they offer insurance to. Customers behave in different ways (different insurance claims patters, different assets insured, etc.). We will soon be able to identify customers who are more likely to submit fraudulent claims and therefore adjust their premiums accordingly. This will enable insurance companies to optimally price their premiums.

Crate Packing Optimisation

Optimising the packing of shipping containers

In many industries, boxes and crates of different dimensions and weight need to be packed in to shipping containers. Packing these containers in an optimal way (so that the minimum number of shipping containers are used) and bearing in mind packing standards (that define various limitations and weight distribution for transport over air, sea, land and air) is a very difficult problem. Isazi Consulting created an algorithm that solves this complex problem. The algorithm finds the optimal way of packing such containers. In one application using this algorithm, our client saves 15 containers per month, with each container costing around R40,000. This means that our client saves about R600,000 per month, or just over R7,000,000 per annum.

Airline Optimisation

Finding the optimum flight schedule for airlines

The airline industry is extremely competitive. Perhaps the most important question for an airline is how to design their flight schedule in such a way as to maximise revenue. Isazi Consulting has developed an end-to-end solution that can solve this problem. Given business intelligence about point-to-point demand between every pair of cities over time, information about how much of the market an airline might be able to capture, and flight constraint and maintenance information, our algorithms are able to design an optimal, consistent flight schedule for the airline. This schedule is optimal in the sense that it generates the maximum possible revenue for the airline. We have also developed tools for changing schedules on the fly (say in the event of a natural disaster or popular concert, when flying demand suddenly changes), and seamlessly stitch schedules together. Our solution, unlike those of other industry players, is provably optimal.

ClassMaster

Data-driven education

Improving education in South Africa was one of the core reasons that Isazi Consulting was founded. We have developed a data-driven solution for education that assists teachers to deliver the best possible education to their students. Where many technology-driven education solutions are student focused, our solution is focused on teachers. Collecting thousands of data points per week about every student, it is able to offer teachers advice on how best to deliver content to the students in their classes. It can also offer essential training to teachers through a mobile device. Backed up by pedagogical theory and advanced software and artificial intelligence, the ClassMaster education solution is poised to revolutionise education across the world, starting in South Africa.

Maize Market Forecasting

Predicting Futures Commodity Prices

Companies who buy and sell maize (or any other commodity) expose themselves to a certain amount of risk because of the uncertainty in the price of maize in the future. The futures exchange provides a way to mitigate this risk, because it allows companies to guarantee a price in the future by performing a 100% hedge. This prevents such companies from making money when the stock prices change, however. We have developed a solution that analyses thousands of indicator variables, including hundreds of relevant weather statistics, political sentiment, financial indicators and past price trends. Our system enables us to forecast the direction of the future stock price of maize with 76% accuracy. This is more accurate than the futures price for the same expiry date, and enables our clients to design trading strategies that can generate far greater revenue in the future than the 100% hedge.

Epidemiological Modelling

Modelling HIV incidence rates

We have done mathematical modelling and built software for an epidemiological modelling company on several occasions. This has included setting up a matrix eigenvalue model to determine the quantity of drug resistant HIV in a blood sample given the results of a PCR (Polymerase Chain Reaction) lab test. We have also built an R package for non-linear mixed effects models where an initial uniform distribution of infection times is added to the model. This was used in a study to infer the incidence rate of HIV from measurements of trace indicators in the blood.

Scene Text Recognition

Finding text in images

There are thousands of problems that involve detecting text in a possibly blurry, low resolution image. Isazi Consulting has developed an engine, usable in the Android operating system, that enables developers to scan text from natural images. This technology is far more advanced than traditional OCR, which requires images to be high-resolution scans of documents. We have developed, as a sample application, an application that allows users to scan airtime slips by photographing them instead of entering 16 digit PIN codes. Another application we have developed, this time in the big-data space, is a tool that can scan through large image databases and extract text from all the images. This can then be used in data queries.