finding extreme values in a panel data set

3 次查看（过去 30 天）
salva
salva 2012-7-22
Dear Matlab experts
I have
A={
[1] ' ' [NaN] [NaN] [NaN]
[1] '30/11/08' [11.1447] [0.2117] [463.1360]
[1] '28/12/08' [11.7129] [0.2209] [436.6316]
[1] '25/01/09' [11.5152] [0.2212] [441.5430]
[1] '22/02/09' [11.4854] [0.2201] [453.7015]
[1] '22/03/09' [NaN] [0.2185] [461.3925]
[1] '19/04/09' [10.9700] [0.2104] [486.4095]
[1] '17/05/09' [11.4846] [0.2162] [451.4833]
[1] '14/06/09' [10.9967] [0.2158] [475.8620]
[1] '12/07/09' [11.1990] [0.2211] [449.4574]
[2] ' ' [NaN] [NaN] [NaN]
[2] '30/11/08' [10.8638] [0.2175] [472.2659]
[2] '28/12/08' [11.0897] [0.2182] [479.3408]
[2] '25/01/09' [11.0996] [0.2286] [442.6719]
[2] '22/02/09' [11.0473] [0.2211] [481.4548]
[2] '22/03/09' [11.1151] [0.2259] [468.2757]
[2] '19/04/09' [10.9865] [0.2300] [461.5581]
[2] '17/05/09' [10.7931] [0.2259] [487.6257]
[2] '14/06/09' [10.5845] [0.2200] [529.8777]
[2] '12/07/09' [10.8926] [0.2315] [433.3039]
[3] ' ' [NaN] [NaN] [NaN]
[3] '30/11/08' [10.7658] [0.2339] [487.7729]
[3] '28/12/08' [10.5227] [0.2305] [505.4553]
[3] '25/01/09' [10.6700] [0.2279] [480.0745]
[3] '22/02/09' [10.7632] [NaN] [518.5076]
[3] '22/03/09' [10.5065] [0.2271] [470.0111]
[3] '19/04/09' [10.6112] [0.2215] [483.5394]
[3] '17/05/09' [10.5855] [0.2190] [497.3172]
[3] '14/06/09' [10.3085] [0.2158] [515.7492]
[3] '12/07/09' [10.4475] [0.2211] [482.3380]
}
A is a panel data set that shows the evolution of some variables (columns 3-5) over time (column 2) for 3 individuals (column 1).
I would like
1) to detect outliers (their position and their corresponding values) in columns 3, 4 and 5 where an outlier is defined as a “a value that it is larger than (or equal to) 3 standard deviations over the mean”.
2) plot a diagram for each of the columns 3, 4 and 5 where outliers will stand out with, say, a red color
Could I do that with some matlab code?
Thank you very much
just to mention that my real A is huge where I have 400 individuals.
  1 个评论
Image Analyst
Image Analyst 2012-7-22
编辑：Image Analyst 2012-7-22
That's far, far from huge. Do you want this done separately for each group listed in column 1? It looks complicated by the fact that not only do NaNs separate the groups listed in column 1, but there are random Nan's scattered about.

请先登录，再进行评论。

请先登录，再回答此问题。

采纳的回答

Image Analyst
Image Analyst 2012-7-22
编辑：Image Analyst 2012-7-22
What have you tried so far? Did you try anything along the lines of
column3 = cell2mat(A(:,3))
goodIndexes = ~isnan(column3)
goodColumn3 = column3(goodIndexes)
mean3 = mean(goodColumn3)
std3 = std(goodColumn3)
lowerLimit = mean3 - std3
upperLimit = mean3 + std3
outliers = column3 < lowerLimit | column3 > upperLimit
outlierIndexes = find(outliers);
normalIndexes = find(~outliers);
x = (1:length(column3))';
stem(x(normalIndexes), column3(normalIndexes), 'b');
hold on;
grid on;
stem(x(outlierIndexes), column3(outlierIndexes), 'r');
% Enlarge figure to full screen.
set(gcf, 'units','normalized','outerposition',[0 0 1 1]);
and so on for the other columns?
  6 个评论
显示 4更早的评论
Image Analyst
Image Analyst 2012-7-22
You first calculate the MAD. Then you multiply that by a factor that says how many of those you need to consider something an outlier, just like you chose 3 for sigma to say that if it were more than 3 sigma away from the mean it's an outlier. Same thing except use MAD instead of the standard deviation sigma. Like, if that data point's value is more than 2 (or 3 or 4 or whatever suits your situation) MADs away from the mean, then that data point is an outlier.

请先登录，再进行评论。

更多回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Descriptive Statistics 的更多信息

标签

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by