Ali Ahmed | Consultant Symbios, Information Systems & Supply Chain Specialist 
This email address is being protected from spambots. You need JavaScript enabled to view it.
Predictive Analytics, a common term we hear nowadays, how this financial stock will perform, who has a high risk of heart disease, which project is most likely to succeed, and so many more questions. In this short article we will get introduced to what it is, how it works (data and algorithms), and some applications.
First predictive analytics is something we all have performed in high school. Remember when we were given x values and had to predict they? typically we started drawing lines in the y and x-axis. This is the most basic form of predictive analytics, which is known as Regression methods. We are all familiar with this basic form of predictions, however, our problems can be more advance where we have more than one x value, where y value is categorized instead of a continuous number, and where our inputs are even text values. This all can result in predictive analytics.
The most common models of predictive analytics are regression, decision trees, and neural networks. We already introduced the former, as for the latter: the decision tree is based on probability trees. Something many of us covered its basics in university, however, this is a more automated form. Typically, it falls under the domain of machine learning, where the machine learns by probabilities which are the most likely outcome. Finally, Neural networks are also a machine learning model where the machine teaches itself which neurons to fire with which weights, each neuron is assigned a weight, thus leading to the right result. The model/machine will automatically test the data with given results and learns which weight to assign the middle part, aka neurons, to get the right result each time. These are the most common prediction models. Neural networks are a mix of regression and decision tree, more complex in nature as it is multi-layered.
So how do we start? Well, a common methodology is known as CRISP-DM. Cross-Industry Standard Process for Data Mining. It is highlighted as being cyclical. See the attached image. First, we need to understand the dataset with the help of a subject matter expert. Second, we prepare the data to fit into our model parameters. Then we fit the model to our inputs leading to our model. We evaluate the model's score using test data different from the original to avoid model overfitting, and finally, in the CRIPS-DM model, we deploy our successful model to be reused intelligently. An extra stage that is required when we have so many AI models running in operation, service, or industry, is to monitor the performance of the model. Each model can rot if it is not maintained and updated.
Popular uses of Predictive Analytics are:
- Finance and banking industry: predicting stock performance a branch of financial engineering, also predicting fraud, and good loan applicants.
- Customer Relationship Management: What the customer is most likely to buy? When might our customers leave us? What offers to win them back?
- Machine failure: aka Preventive Maintenance where we based on data like performance sensor data, and output rate we can predict the lifetime of the machines.
- Energy consumption: based on demand, weather, and historical records we can predict future consumption.
- Money gathering campaigns: for example, collection agencies and humanitarian agencies can predict who will pay, which channel will make them more likely to pay, and what tools to use to make them pay.
Other less common uses of predictive analytics, where we still depend on experts’ judgments are:
- Health predictions
- Autonomous driving or driver aided driving.
- Risk management
- Child protection services
We hope by the end of this article you have gathered a clearer idea of Predictive Analytics.
For questions and comments please email @ This email address is being protected from spambots. You need JavaScript enabled to view it.
See Blog on Symbios Digital
