Welcome to the final part of my 3-blog series on building a predictive excellence engine. Accurate forecasts have the potential to accelerate a firm’s growth as they impact strategic business decisions such as sales targets, stock value, financial planning, raw material procurement, manufacturing and even the supply chain decisions. Though guesstimates and traditional time series analysis are the popular approaches for forecasting, emergence of artificial intelligence necessitates and facilitates a re-thinking of old practices to improve accuracy.

This blog series describes an innovative forecasting approach which is able to overcome the challenges associated with forecasting by seamlessly merging time series analysis with artificial intelligence (AI).Forecasting essentially comes under time series analysis. But artificial intelligence can be a force multiplier in terms of heightening the accuracy of forecasting as it has the capacity to study and discover fascinating patterns in the data automatically. There are two key ingredients in our forecasting recipe:

1. Intelligently engineering and identifying key drivers of the target variable
2. Using these drivers in an ensemble modelling approach consisting of successful time series and AI models

The first part of the blog series gave a brief introduction to time series analysis and the tools needed to makes sense of such datasets (Link here) while the second part focused on feature engineering and selection (Link here). We will be deep-diving into the final predictive engine in our driver based forecasting approach in this blog. It is made up of 10 models (5 time series and 5 AI models) and in the end we take average of all the model predictions as the final forecast of our predictive engine. Given the large number of models, we will give a brief introduction to each along with details on how to implement them in python. In a separate blog we will discuss the best practices on optimizing each of these models.

The dataset being used here is the same as the previous blogs in this series. It is a daily dataset on Hong Kong flat prices along with 12 macro economics variables. The target variable to predict is ‘Private Domestic (Price Index)’ i.e Hong Kong flat prices. As highlighted in the previous blog, we chose a set of 3 drivers of Hong Kong flat prices:

1. Total stock
2. M3 (HK$ million)
3. Lag 1 of M3 (HK$ million)

#python #time-series-analysis #ensemble-learning #forecasting #ai