Wolfram Language Paclet Repository
Community-contributed installable additions to the Wolfram Language
Outlier detection for financial time series
XXXX.
Data
Here is a time series and its length:
In[1]:=
tsData=FinancialData["GE",{{2019,1,1},{2024,1,1},"Week"}];Length[tsData["Path"]]
Out[2]=
261
Here is a date list plot with a vertical grid line that marks the training and testing split date:
In[15]:=
splitDate=DateObject[{2022,1,1}];pl=DateListPlot[tsData,AspectRatio1/3,ImageSizeLarge,PlotLabel"GE stock",GridLines{{splitDate},Automatic}]
Out[16]=
Via geometric nearest neighbors
Here we split the data into training and testing parts:
In[17]:=
splitPos=Length@Select[tsData["Times"],#≤AbsoluteTime[splitDate]&];{training,testing}=TakeDrop[QuantityMagnitude[tsData["Path"]],splitPos];Length/@{training,testing}
Out[19]=
{157,104}
Here we get the positions of the anomalies:
In[20]:=
pos=
[training〚All,2〛,testing〚All,2〛,"AnomalyPositions","NumberOfNearestNeighbors"10];Shallow[pos]
 | GNNMonAnomalyDetection |
Out[21]//Shallow=
{1,2,3,4,5,10,11,12,13,14,50}
Here is a plot of the time series and the found outliers:
In[22]:=
Append[pl,Epilog{Red,PointSize[0.012],Point[testing〚pos〛]}]
Out[22]=
 |
|
""