Anomalies and fraud detection
Numerous companies are facing the problem of customer fraud and loss of money due to misrepresentation. However, many of these companies have large amounts of data about their customers due to the nature of their activities. The effective use of this data can allow businesses to quickly identify potential cases of fraud without having to devote significant resources to manually reviewing files. In one case we helped a client to resolve problems with fraudulent customers. These customers were providing false meter data and were therefore underpaying for their consumption of energy.
DIVERSITI endeavors to provide clients not only with a list of potential fraud instances, but also a ranking of all clients in order of likelihood of fraud. This enables our clients to utilize their fraud detection resources in the most efficient manner possible by only looking at the most likely cases. DIVERSITI applied a combination of dynamic time warping and anomaly detection algorithms to generate a probability of fraud for each customer in the dataset based solely on the time-stamped data of their energy meter. The anomaly detection algorithms work by identifying unanticipated changes in the Energy dataset and dynamic time warping was used to identify similarities with previous customers. Thus, the DIVERSITI approach allowed customers to be compared to their own past behavior and to other past cases in order to identify fraud.
Several hundred suspicious meter readings were detected, allowing the company to inspect the most likely fraudulent customers. This automated analysis saved thousands of man-hours and substantially reduced fraudulent underpayments.
There is a risk of fraud and meter breakage for nearly every business. The advantages of this type of analysis are not limited to detecting anomalies in time series data alone, we can apply very similar patterns to credit card spending, insurance claims and any type of prorated billing. In each case, companies can derive probabilities of fraudulent activity for each customer and, from there, they have a great deal of flexibility in how they want to use the data. They can only look at the few most likely cases of fraud or can work their way down the list. No matter how many clients they choose to review manually, they can always be sure that they are using their resources as efficiently as possible by examining only the most likely instances.
Object Detection Using Live Video Feeds
Many businesses have security systems in place to protect their workers and valuables, with a common practice being security cameras inside and outside a facility. These cameras stream live video feed during the entire day, and sometimes at night depending on the business’s needs. Security cameras normally require a person to monitor the live feed, however computers can be taught to perform that task as well.
Object recognition is computationally demanding problem for a machine. The majority of the issue lies in the variation of an image, since any object can create an infinite number of different images based on its position, orientation, size and lighting. Identifying a car in a parking lot is a simple task for a human being that becomes significantly more difficult for a machine. The real-time aspect of a video, essentially many sequential images, means that any object detection algorithm used must execute almost instantaneously.
Diversiti’s Approach was to develop software that swiftly and accurately classified the objects in a video. We needed to detect several moving objects at once, so we isolated the objects of interest by removing the extraneous background information. A convolutional neural network was chosen as our model due to its superior capabilities for modelling image recognition. Potential scenarios included car crashes and people fighting, and was generalized to cover a variety of situations. After training and parameter tuning, this model could systematically detect and tag important objects. The speed of the neural network model allowed for the recognition of objects on a real-time video feed.
This automated system for tagging objects greatly reduces the need for constant video monitoring. Without the need for a human camera monitor, thousands of work hours can be saved. When the software detects an emergency scenario, and is certain of it, the authorities are automatically alerted to the situation. Otherwise it saves its findings to be reviewed by a human operator at a later point in time. As an added security measure, the facial recognition aspect of this software can populate a database of recurring visitors and use that information in its object tagging. DIVERSITI provides a tool that not only accurately detects and reports important scenarios, but improves itself over time to better suit its environment.
Automated Classification of Images
A huge number of high quality photographs are being taken in the smartphone era. For many companies there is value locked within these images but the vast number of such images makes them tedious to sort through. Identifying features and objects in these images automatically allows companies to gather information in new ways and while expending fewer resources. This can be used both with publicly available images and with user supplied images, allowing companies to get trained eye on something without actually paying a trained employee to look at it.
One of our clients who need help in this area was a construction company who wanted to use amateur photos of houses to learn information about the layout and dimension of houses without having to send crews to the site.
Our data experts is able to combine advanced machine learning algorithms to suit our clients’ needs. Machine vision is a cutting edge field and there are many different algorithms each with strengths and weaknesses in different tasks. In this case we combined the types of algorithms used to create panorama images with convolutional neural nets for object recognition. This allowed us to create an accurate 3d model of a house using only photos of the outside.
The ability to have a computer instead of a person extract relevant information from an image can save companies a great deal of money, especially if the necessary identification requires a great deal of domain knowledge. Trained humans are expensive but once you train them computers are cheap. In this case our algorithm allowed the construction company to forgo sending employees out to properties to collect basic information.
Ready to get started?
Our experts and software engineers have worked with industry managers, collecting and analyzing information to transform business processes. We’d like to help you create your next successful solution, whether it’s migrating existing platforms or developing new systems. Technology is evolving faster, let’s build sustainable solutions.