0
$\begingroup$

I have past 1 year of sales data and now i want to predict sales for next coming year. How can i predict it.

$\endgroup$
  • $\begingroup$ You have one year of sales data, and want to predict another? Not possible. You'd need a set of training data (comparable prodocts, comparable distribution, comparable region, comparable point in life-cycle, comparable everything) with thousands of years (accumulated) of Sales to train a model that would work with any kind of confidence. - I'd go with a NI-approach, aka natural intelligence, aka a consultant with deep ties to, and long experience in, the industry in question. $\endgroup$ – bukwyrm Feb 27 '19 at 10:24
  • $\begingroup$ voted to close as too broad. $\endgroup$ – Joe S Feb 27 '19 at 13:32
1
$\begingroup$

You can not predict sales for the whole next year by training your model with only one year data.

But still one year data is better than having nothing. And you can analyze one year data to find trends and patterns. On top of those trends and patterns, you can apply domain knowledge and make some very some useful strategic decisions.

Problem you are trying to solve is called Time Series Forecasting as temporal order of input matters while training.

Here are reference points you can your start with :
1) https://machinelearningmastery.com/how-to-get-started-with-deep-learning-for-time-series-forecasting-7-day-mini-course/
2) https://www.analyticsvidhya.com/blog/2018/02/time-series-forecasting-methods/
3) https://towardsdatascience.com/an-end-to-end-project-on-time-series-analysis-and-forecasting-with-python-4835e6bf050b

|improve this answer|||||
$\endgroup$
0
$\begingroup$

You can also try random walk models with confidence intervals. The idea of random walk process is it tries to simulate what might be your next point based on your current point. Random walk models are of many types, particularly in retail (sales data) you usually come across 2 types:

  • random walk with drift (drift will be an upward or downward trend in your data)
  • random walk without drift

The forecast package in R offers an excellent method auto.arima that you can use to implement this technique. If auto.arima is not able to fit an ARIMA model on your data it usually fits an ARIMA(0,1,0) which is tantamount to random walk.

You can find more information about different types of ARIMA processes here and random walk process here.

|improve this answer|||||
$\endgroup$
  • $\begingroup$ Could you expand on this? Say he has data: 2018January:5M€; 2018February:5.1M€ ... 2018December:4.7M€ --- would the 'random walk' do anything else than say: 2019January = 2018December +/- Deviation(2018) = 4.7M€ +/- 0.2M€? $\endgroup$ – bukwyrm Feb 27 '19 at 14:06
  • $\begingroup$ Yes I believe that is true. $\endgroup$ – Prathmesh Savale Feb 27 '19 at 14:24
0
$\begingroup$

I recommend using Facebook’s open source tool for forecasting named “Prophet”: https://facebook.github.io/prophet/docs/quick_start.html

|improve this answer|||||
$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.