Behavioral data
| Behavioral data |
|---|
| See also |
Behavioral data is gained as output from behavioral experiment, collected from individuals or groups [1]. It is measurement of a behaviour or action. It is collected to predict future, for example to estimate how consumer might behave or react in the future, but also to see the customer's characteristics right now, for example to estimate how loyal the customer is. It answers question: what customer do when he interact with the business? by recording his actions [2]. Depending on which industry behavioral data is collected, it may refer to
[3]:
- amount of sales,
- dates of purchases,
- types of purchases,
- dates of payments,
- dates of purchases,
- customer service activities,
- claims of insurances,
- behaviour of bancrupcy,
- webside activity,
- sales on websides,
- path which user takes in webside,
- other.
Data types and charactericts
Behavioral data is considered as the most predictive type of data, however it is the most expensive one and the most difficult to gain if the company wants to get it from outside source. If company has access to its behavioral data, which means knowledge about operations by product, it is available at zero cost [4]. Other types are demographic data and psychographic data[5].
| Type of data | Power of prediction | Stability | Cost |
|---|---|---|---|
| Demographic | Medium | High | Low |
| Behavioral | High | Low | High |
| Psychographic | Medium | Medium | High |
Challenges in modeling behavioral data
Fennell P. G. points several reasons why behavioral data is challenging to model [6]:
- scale - of research, of data and size of final database,
- sparseness - because of several observations per individual,
- heterogeneity - because individuals behave differently,
- class imbalance - several observations of the outcome of interest,
- interpration of models - required some time to learn it, usually there are many complex algorithms to predict behaviour of potencial customer.
Tools of behavioural data analysis
Behavioral data is usually huge size database, there are several approaches used in analyzing. Example of ready to use model is structured sum-of-squares decomposition (S3D). Such model includes [7]:
- algorithms,
- decision trees,
- variant of regression trees,
- variation of the outcome,
- correlations between the features,
- statistical models,
- support vector methods.
Footnotes
References
- Bari A., Chaouchi M., Jung T. (2016), Predictive Analytics For Dummies, John Wiley & Sons
- Fennell P. G., Zuo Z., Lerman K., (2018),Predicting and Explaining Behavioral Data with Structured Feature Space Decomposition
- Gerber L. R., (2006), Including behavioral data in demographic models improves estimates of population viability, in "Front Ecol Environ 2006; 4(8)", The Ecological Society of America, Arizona State University
- Hartmann P. H., Gotrman J. M., Jones R. R., Gardner W., Kazdin A. E., Vaught R. S. (1980), Interrupted time-series analysis and its application to behavioral data in "Journal of applied behaviour analysis 133, number 4 (winter 1980)"
- Rud O. P. (2001), Data Mining Cookbook: Modeling Data for Marketing, Risk, and Customer Relationship Management, John Wiley & Sons
- Turner M. B., Forstmann B. U., Steyvers M. (2019), Joint Models of Neural and Behavioral Data. Computational Approaches to Cognition and Perception, Springer
Author: Aleksandra Puchała
