How to forecast monthly time when we have only one year data?

data_science

#1

I’ve a Employee Absenteeism data which has 21 variables and 740 rows. The objective of the study is: How much losses every month can we project in 2011 if same trend of absenteeism continues?

I have grouped the data based on month to get the total absent time for that month. The below is the data frame.

      Category        x
1         1       177.6050
2         2       276.1621
3         3       458.7209
4         4       238.6340
5         5       266.1499
6         6       243.5155
7         7       376.9841
8         8       250.3904
9         9       182.4135
10       10       293.2177
11       11       267.1706
12       12       193.9427

When I’m fitting this data on time series linear model I get this:

data = ts(aggre.absent.hours.months$x, frequency = 12, start = 2010)

fit = tslm(data ~ trend + season)
summary(fit)
tslm(formula = data ~ trend + season)

Residuals:
ALL 12 residuals are 0: no residual degrees of freedom!

Coefficients: (1 not defined because of singularities)
            Estimate Std. Error t value Pr(>|t|)
(Intercept)  176.120         NA      NA       NA
trend          1.485         NA      NA       NA
season2       97.072         NA      NA       NA
season3      278.145         NA      NA       NA
season4       56.573         NA      NA       NA
season5       82.604         NA      NA       NA
season6       58.484         NA      NA       NA
season7      190.468         NA      NA       NA
season8       62.389         NA      NA       NA
season9       -7.073         NA      NA       NA
season10     102.245         NA      NA       NA
season11      74.713         NA      NA       NA
season12          NA         NA      NA       NA

I know the reason why linear regression gives NA as we don’t have enough samples(rows) as compared to the predicted variable.

Question is how can we forecast the monthly absent time when we have only one year data with trend? Am I missing something here.
Data Set.zip (20.4 KB)

Attached original data set.

Your help will me much appreciated! Thanks in advance!


#2

12 datapoints are just not enough to suggest a model. Moreover, there could be an yearly seasonality in the data. We need atleast 2 years of data to capture any yearly seasonality.


#3

I’m concerned about monthly trend applied to the next year as this is the objective(if same trend continues how much losses every month)


#4

You can use the same value as predictions (kind of naive method). So for January 2011, the number of hours would be 170. Similarly, for February it will be 276 …