Signals empower us to predict the future by learning from the past.
We all learn from the past. So if you failed a mathematics exam in school, you learned you had to study harder not to fail the next one. Events that happened in the past can be measured in many different ways. Measuring a past event puts it into focus. In data mining, this process translates into creating a descriptive feature to describe or tell a story about the past in some way, shape, or form.
We call these descriptive features Signals. Signals are the indicators extracted from raw data that have proven to be valuable for solving a particular problem. Signals can also be created from other Signals by transforming one piece of information into something more meaningful or interesting. For example, using the initial Signal extracted from raw data — in this case, a test score — we can create several Signals that capture past school performance in mathematics and science for a certain student.
For example:
Descriptive Signal 1: Grade on the latest mathematics exam. Note that in this case, we are binning an initial raw score, say from 0 to 100, into bins that we believe are more meaningful in measuring performance. In this case, we are transforming the raw score into A, B, C, D, or F, depending on a set of pre-selected intervals. The event here was the score associated with the last mathematics exam, and the transformation was the process of binning the score into letter grades.
Descriptive Signal 2: Average score on the latest math and science exams. Note that we are using an average to combine several raw input fields, the score from the last physics exam, the last calculus exam, and so on.
Descriptive Signal 3: Average score in mathematics exams for the last six months. Note that this Signal tells a lot about performance over time. Signals that aggregate values in such a way are very powerful. Even if the first two Signals point to poor performance in science, the third Signal can tell a completely different story since it measures the student’s overall performance based on a number of events going back in time. It may be that he or she got seriously involved in a school project (e.g. glee club) for a week or two and had no time to study for a given exam.
Whenever we determine that we have enough Signals to understand what happened in the past, we can use them to predict what will happen next.
Intelligent Signals
The future is always more difficult to measure than the past. Nonetheless, we commonly use the past to predict the future in our daily lives. For example, based on the descriptive Signals we derived above, we can use past school performance (Signals 1, 2, and 3) as input into a model to help predict what will happen next (e.g. propensity to fail in the next mathematics exam). And if such a model determines that poor past performance indicates a high probability of failing in the next exam, we may recommend a strategy that will increase the student's chances of passing the exam the next time around. Note here that we are using descriptive Signals as inputs to a predictive analytic model for which the output is then used as input into a prescriptive model. In Opera Solutions' Signal Hub, both predictive and prescriptive models are also Signals. We call these intelligent Signals, though, since they forecast an outcome or behavior as well as provide a decision in order to mitigate or augment the chances that something will happen.
Predictive Signal: The propensity that this student will fail the next science exam. Predictive models can be defined as clever mathematics applied to data. They are able to discover patterns in data (obvious or hidden), learn when these patterns occur, and use them to predict a certain outcome. In Signal Hub, predictive models are intelligent Signals that can be used for a variety of business cases or scenarios throughout the enterprise.
Prescriptive Signal: Advice to give a student once we know his/her propensity to fail. Note that this is how the result of a predictive Signal gets translated into action. That is, the Signal is used to select the right decision or approach given the requirements, costs, or constraints around the problem we are trying to solve. For example, if the propensity to fail in science is very high, we can decide to bring in a tutor to help the student. If the propensity is medium-high, we may decide to suggest more studying hours.
Descriptive, predictive, and prescriptive Signals come together in what we call the Signal Layer inside Signal Hub. Through the Signal Layer, Signals are shared and re-used across the entire organization to generate value and insight as well as to address a myriad of use cases in a fast and concise way.
How Are Signals Generated?
Signal Hub makes it extremely easy for data scientists and users to implement descriptive and intelligent Signals. It offers a smart environment in which Signals are created, maintained, and shared across the enterprise.
Want to learn about how Signals can help predict future behavior? Check out our white paper on transformative analytics:
