Complex stuff! However, we can start from the basics , gain confidence in understanding our dataset and utilise whatever has already been done by other stalwarts!
We have a 11 years monthly data on milk procurement by more than 100 milk co-ops spread across India. This contains their geographic location, quality of milk procured, price received , products made etc etc. This is a seasonal industry in India affected by weather, prices and in general crafty management acumen on timing of buy and sell of milk commodities.
First, we initiate the time series data mining without looking at the location. We pick the low lying fruits viz. DTW, Euclidean, Mahalanobis distance matrix using R codes. Then,outliers from tso, tsoutliers packages in R and also the anomaly detection VEC package of Twitter. So far so good, many interesting results have been thrown out but more you see, more you enter into deeper waters...what was the cause for all that? We are now cross checking with the organisations and believe me , we are getting tremendous insights into the business..like a detective story!