Ensemble Sales Forecasting Study in Semiconductor Industry

---

This democode has been tested under R version 3.3.2 (2016-10-31) -- "Sincere Pumpkin Patch"

If you have questions about the use of this package, feel free to contact Qiuping Xu: qx0731 AT gmail DOT com

This is the supplementary material for paper: Q.Xu and V.Sharma,'Ensemble Sales Forecasting Study in Semiconductor Industry'. Appears in Industrial Conference on Data Mining, 2017. arxiv version is at https://arxiv.org/abs/1705.00003.

You are welcome to use the democode. If you use it for a publication, we would appreciate an acknowledgement by referencing our paper.

These scripts were developed for generating ensemble forecast model for time-series type of data. For iid data, the variable importance is automatically muted. 

These algorithms were developed for:

1.decorrelation (derorrelate.r) of the features by clustering feature into groups and select the representative feature of each group.

decorrelate = function(data, k1, k2)

	# input: 
	
  # data: data frame with the first column as the response and the rest are the features 
  
  # k1 the dimension for MDS
  
  # k2 the dimension for Kmean
  
	# output: 
	
  # list of 3: data frame, mds fit,  kmean fit

2.Ensembled linear regression methods and variable importance extraction (linear_source.r)

LM_ensemble_ts_para = function(target, target_test, nmodel, Kmax, test_number, time_sample, rank)

	# input: 
	
  # target: traning data
  
  # target_test: test data set
  
  # nmodel: number of models to save for changepoint detection 
  
  # Kmax: max number of features for model building 
  
  # test_number: number of sample in validation set
  
  # time_sample: Boolean whether the data is a time series data
  
  # rank: Boolean for variable importance rank. Mute for iid data ( time_sample is FALSE)
  
	# output: 
	
  # list of 3
  
  # item 1: frct value
  
  # item 2: frct [low, mean, high] for each model in the ensemble 
  
  # item 3: variable rank

3.Ensembled random forest methods and variable importance extraction (RF_source.r)
	
4.Ensembled Extreme Gradient Boosting methods and variable importance extraction (boosting_source.r)

5.A demo on a toy dataset (demo.r)

Due to the sensitivity of the original data used in the study, the democode was based on simulated datasets. The toy data set contains 11 features (8 numerical and 3 categorical). The response variable is a function of f_1 and f_other1 (response = 0.1* f1 + f_other[,1] + white noise).

Key outputs:

• Line 49 in 'demo.r': see whether 'f_1' and 'f_1_similar' were clustered into same group

• Line 64 to 67 in 'demo.r': see whether the ensemble process gave a reasonable frct number.

• Line 70 to 73 in 'demo.r': see whether the variable ranking correctly identified f1 and f_other1 as the most importance variables. These should be also available in the graphical outputs.

Enjoy ~