Please explain which one would be better to understand the underlying principles in data, Statistical or Algorithmic Modelling?
@Pankaj_Rana The step of modeling itself comes after you have gained a reasonable understanding of data. i.e. each of the variables independently, in relation to each other and in relation to the predicted variable. So, as explained in the myths of data science, we can’t expect to throw data at tools and hope to get ready insights.
The difference as emphasized in the course between statistical and algorithmic modeling is that a statistical model is more transparent and allows you to better answer how the model arrived at a certain answer for a given set of inputs. We can’t say the same about algorithmic modeling.
In lecture : It has been taught as “The step of modelling comes after the analysis.”
Collect -> Store -> Process -> Analysis -> Modelling.
In reality, this entire process is an iterative one. There might be a possible scenario that even though your data analysis and your model are beyond perfection but the data you are using might not be right.
Perfect your EDA skills. This will significantly help you in your modelling stage. Exploring simpler relation between the features should be done first, but if the relation isn’t linear, quad or even cubic, it’s favorable to opt for algorithmic modelling.
Make sure you have enough data for the training.
Hope this helps. Cheers!
Thanks for the answer!!! It has somewhat clear my doubt…
But what will be the answer of the question below…??
If some client ask me to explain the relationships between the parameters of a machine to predict the state of a machine(going good or about to crash) and the data available is huge like humidity, temp, rpm, running hours etc which have been captured for 5 years everyday every hour. The client wants it so that he can take care of situations like if humidity is increasing, he needs to reduce the rpms and combinations like that which the client to able to judge manually or by simple graphs. He wants to know why did the machine crash.?? So should I go for statistical modelling or algorithmic modelling for prediction??
Thanks for the answer…!!!
So, does it mean that when relationship is not linear… I cannot explain my client as why something is happening??
As far as prediction is concerned as per lecture, algorithmic modelling answers that. Statistical modelling more deals with drawing insights from present data, and expressing it in linear or normalised form.