When to use modeling in data science.

So recently I had the fantastic opportunity of meeting up and engaging with Carey Nadeau Co-Founder and Co-Ceo of Loop. They are an awesome platform doing amazing stuff in the insurance space so please be sure to check them out!

And although the conversation, in general, was pretty cordial, jovial, and productive at best, Carey asked me this question- ‘When do you need to use modeling on a data science project’, and it was only up till this point that I had an ‘uh-uh’ moment as I had actually never really thought of that.

So today, I will be narrating my somewhat ‘generic’, but hopefully, useful response to that question, and here it goes!

So when asked the question my response was, I feel modeling in a data science project has to be considered based on a couple of factors:

Firstly, what is the problem at hand that we are trying to solve?

Understanding the problem is one of the most important components in deciding ways in which to tackle the problem. For instance, is it an insight-deduction-based problem, is it a predictive-based problem, etc. As such not all data science projects have to necessarily end in modeling.

Consequently, let’s say we have a potential client who wants to explore and understand a certain business/industry’s market and profit trends so as to help them determine if this is a potentially lucrative investment, such a project would rely heavily on us exploring and deducing insights from the already available data (historical data).

Hence such a problem would be aligned with data analytics and visualizations as we are using already given data to advise and give insights, there is most likely no algorithm prediction needed.

However if it was more of a question of a business wanting to introduce a new product and need advice on whether this product would be well received, such a project involves having predictive analysis hence ideally a model would have to be built in order to aid in making a data-backed prediction.

More often than not, modeling works when we are trying to make use of numerical data in order to determine numerical insights to support or promote a certain decision. This explains why in order for data features to be useful in model development we need to convert them to numerical weights.

Moving on, based on my prior project experience modeling is more relied on when working with data science projects where we are seeking to automate a system. This is why industries such as trading, exchanges, or internal business operations tend to rely more on models in order to aid in giving backing to a decision.

This may be seen in how a quote is made for a potential insurance customer, how economists can predict currency performance or businesses may analyze month on month forecasts. Modeling is suited for more repetitive and long-term tasks that require a somewhat ‘once and for al’ type of fix — hence just code it!

So in a nutshell, that was my quick response to the question and hopefully, you found this not only useful but insightful and informative as much as I did!