The closure the analytics is to reality, the higher is the impact
Medium-level descriptive analytics with a higher correlation with business scenarios is far more valuable than high-level predictive analytics with a weaker correlation with actual business scenarios
Classical machine learning algorithms have been used in solving multiple real-life-based problems. Even one of the most fundamental algorithms like linear regression has been used exhaustively in sales and revenue forecasting, marketing analytics, and demand forecasting. The power of analytics is massive when it is used for a given business scenario.
In my last project, in the domain of political consultancy in West Bengal, we could easily identify a region from where responses have increased drastically. Upon studying the basic parameter like text frequency of grievances, we could identify that most complaints were political and were not related to governance issues. Now strategy to influence people in favor of the client will ensure that the political aspirations of the people are fulfilled. Beyond a point in data, a domain expert can decide the due course of action based on other factors as well. However, there were some regions where governance issues were more prevailing highlighting the fact that people are expecting the implementation of policies on the ground. Patterns among gender, geographical regions, and the rural-urban area had a decently identifiable pattern.
Now can’t I just analyze previous election results for each Assembly Constituency and predict the future estimates for votes and vote share? Won’t that be much useful? Yes, of course, results will be able to reflect the indicative voting patterns. But then there are some regions where more than party, a candidate speaks more about people’s preferences. Like in Meghalaya, individuals played great roles, especially in Khasi and Jaintia regions. Other factors like division/fractioning of a political party, delimitation of constituencies, quality of governance, or even a political event can have a great impact on how people’s inclination will change towards a particular party.
But how can we then use data analytics to gauge people’s preferences? Since elections have a heavy influence on real-life incidences, identifying and analyzing long-term choices can be more helpful.
More than user analysis based on the first interaction, engagement over some time can be more reliable.
In this industry, crude or traditional voters are going to be more predictable. Now we can dig dive deeper into linear regression as well keeping different trends in mind. The point is, a data person should not just be concerned about how an error is going to be estimated. Of course, possible deviations, outliers, and errors are crucial to a project. But if one understands the nitty-gritty of the actual business problems, identifying the right methodology for insight will become more beneficial.
Another example, I would like to quote is based on simple correlation and dependency among two factors. Let’s assume that through some measures we can study the historical vote pattern of a person. Along with that, we can also roughly judge her present inclination, then how many kinds of analysis can be done from this? More often, people will deep dive into historical patterns. One will also deep dive into present trends.
But what if we connect both? For example, if I had chosen X in the last three elections and still I would like Y to come in the next election. Then it says a lot about the failure of X in fulfilling the promises. Especially if the percentage of such people is high, despite many numbers of followers, there is a high chance that X will not perform well. Therefore reading between the lines by using dependencies among these parameters is going to be more useful. Now in such scenarios, studying normalization techniques in detail for only one question might not be useful. First of all finalization of the normalization methodology is not easy due to multiple uncertainties. Therefore before deep diving into statistical factors, a basic correlation might have a significant impact.
Nevertheless, mathematical and statistical concepts have a huge importance in studying data. The success of any analytics or algorithm will depend on how good is the understanding of the business scenario. A basic concept like standard deviation can tell patterns within data by studying the spread across and in the depth of a parameter.
The closure the analytics is to reality, the higher is the impact.
Therefore, medium-level descriptive analytics with a higher correlation with business scenarios is far more valuable than high-level predictive analytics with a weaker correlation with actual business scenarios.
