Analytics project: Predictive Modelling and Forecasting
Predictive modeling process uses data mining and probability to forecast outcomes and has extensively usage. This is a complex process which incorporates data analysis, statistical algorithms and machine-learning techniques. Predictive modelling consists of three main steps:
- Building a model or several models, when necessary;
- Model/s Testing;
- Model/s Validation and Evaluation;
For each of these steps, several ways of implementation are possible. From our experience, we have learnt that the best selection depends on the specific problem, the time consumption, human resource and data availability.
Our goal is to go beyond descriptive statistics and reporting and to provide the best assessment on what will happen in the future in order to improve the client decision making process and to produce new insights that lead to better actions.
We found out that clients are mainly interested in predictive modelling in order to:
- Predict trends;
- Predict behavior;
- Improve business performance;
- Drive strategic decision making;
Predictive models are very usefull in terms of:
- Security and Fraud Detection – where predictive analytics can help stop losses due to fraudulent activity before they occur.
- Operations – where predictive analytics is important because allows many organizations to function smoothly and efficiently. They usually target price adjustment, better resource management, asset management, revenue maximization and many others.
- Risk Management – where one of the most common usage of predictive analytics is in credit scoring. Credit scores are used to assess a buyer’s likelihood of default and to other risk-related uses, including claims and collections. The predictive models are used to generate customer credit score, in other words a number, that incorporates all of the data relevant to a person’s credit-worthiness.
In our projects using different software, we have used various predictive analytics algorithm solutions to build efficient predictive models, some of them are:
- Time series algorithms – single, double and triple exponential smoothing, which perform time based predictions;
- Regression algorithms – Linear, Exponential, Logarithmic, Geometric and Multiple Linear regressions, which predict continuous variable based on the other variables in the dataset;
- Survival analysis – which is analysis of the time to events;
- Factor analysis – Maximum likelihood algorithm;
- Decision trees algorithms – predict one or more discrete variables based on the other variables in the dataset;
- Outlier detection algorithms – which detect the outlying values in the the dataset;
- Introduction to Credit Risk part of the Master’s Programme in Applied Econometrics and Economic Modeling.
- Credit Risk Management Course part of the Master’s Programme in Applied Econometrics and Economic Modeling
- 2-Day Workshop Humans and the Art of Analytical Modelling
- CONFERENCE GRANT AWARD
- Credit Scoring and Credit Control XVI