What are the reasons behind price fluctuations? What other data is correlated to price fluctuations? about humanitas HOT 9 CLOSED

TODOs:

• Detect seasonal effects
• find out harvest time of selected commodities and include weather data at the time of last respective harvest in prediction?

from humanitas.

@mstefanro @ChingChia @Fabbrix for this issue it might be wise to perform a meta-analysis of all our results to conclude seasonal effects. What do you guys think?

from humanitas.

@albu89 Sure, could you explain the results of the association rule analysis in more detail first though?

from humanitas.

@Fabrix are you interested in what the ouput says?
0.761 0.981 BajraFAQ=big MeatMutton=small -> MaizeFAQ=big tells us that this implication has a 0.761 support meaning that Bajara and Maize have a big price difference with respect to the previous week where Mutton a small price difference. This can be observed over all years every week with a probability of 76.1 percent. A confidence of 98.1 tells us that if Bajra shows a big price difference and Mutton a small one then with a 98.1 probability this implies that Maize will endure a big price difference.

from humanitas.

sounds interesting, but how do we filter out useful rules? Just by skimming through the rules they looked to me more like correlations lacking explanatory power than rules that actually encode a causal relationship

from humanitas.

The causal relationship is expressed with the confidence -> Confidence = P(consequence | antecedent) where a rule has the following syntax: Antecedent -> Consequence. I defined the usefullness (filtering) of the rules according to the two metrics confidence and support. I think if we put the rules in context it has a huge expressive power. In particular with regards to the humanitarian aspect of our project. The rules could be used in form of a recommendation system. I.e. a user can filter for date and region and get purchasing recommendation. -> {Delhi will experience a medium price increase of rice as suppose to wheat that ist very likely to see a big price decrease during Febuary.} Did this answer your question?

from humanitas.

Yeah, but you computed the associations over all the data right? So all you can say is from the data checked we can say with confidence X any given time that the price of commodity Y will (increase/decrease) by a (small/large) amount if the following rule .. applies to commodity Z

from humanitas.

no the associations are performed from the data of a specific city so we can make regional conclusions. i could also try to include a further filter if you want to narrow it down to a specific season (summer, winter..) what the rules tell us is that if a commodity X is in state x2 then commodity Y will be in state y1. Furthermore we know from the support that out of all the posible states this specific combination (X =x2 -> Y=y1) will occure in Z% of the times. We therefor can not only conclude the relationships but also say in what state commodity X will most likely be.

from humanitas.

Use

• Temperature (offset by 'time to market'/delivery time for commodity)
• Rainfall (significance for different commodities? wheat vs rice)
• Inflation as a price driver (computed from CPI)

from humanitas.