Wondering what makes SiteZeus unique?
Linear regression technique
Interpretability is one of the core benefits to simple forms of statistics. The diagram below represents how a linear regression model works. In principle it involves multiplication and addition. This allows the user to know exactly what variables are affecting all projections and how. This is full interpretability:
The fundamental drawback is the subpar predictive power, driven by interpretability itself. The exact same variables and weights (coefficients) are used on any and all projections. With this limitation, linear regression cannot deal with conditional situations, nor does it become flexible in uncommon situations. For example, in real estate, different geographical markets are driven by different factors that ultimately lead to a business’s success.
Linear regression has severe limitations in delivering this insight into projections. In this scenario, naturally, any projection will be an average effect of all markets involved in the sample set leading to subpar results. Modern machine learning fundamentally builds using conditional. statements (decision trees), leading to more predictive power.statements (decision trees), leading to more predictive power.
Modern machine learning
Random forest technique
Accuracy and predictive power of modern machine learning methods like random forests, neural networks and various other iterative, ensemble algorithms have the benefit of mitigating overfitting (which means the models over compute to the noise in the data and ultimately have far less true predictive power in real world application).
Modern machine learning can build unbiased models by sampling subsets of the data set to build “multiple models” inside of oneself, ultimately creating a decision network or tree (shown below) that simulates reality in a truer sense, all the while resulting in better overall results through statistical averaging.
Interpretability of complex algorithms and modern forms of artificial intelligence is the trade off to predictive power and enhanced accuracy of models. This means for any given forecast or projection, the user will not fully under- stand every factor affecting the overall outcome or result of a given projection. For example, Random Forests builds hundreds of decision trees that ultimately average to a single projection. Trust must be built on cross-validated results.