Algorithm 1. The Proposed Model (XGBoost + MRMR)

Model Inputs:

N: Number of power outage instances

F: Features (F1, F2, …, F26)

Model Output:

T: Power outage duration (t1, t2, …, tN)

Begin:

//Data Collection

for i = 1 to N do

record Data i    //insert power outage data instances

end

//Data Preprocessing

Analyze the Data //Feature Engineering

//Feature Selection

for each feature Fi in Data do    //each feature Fi in Features

score (Fi) = CalculateMRMR (Fi)    //ascertain feature importance with MRMR

end

MS = max (score (Fi))    //select features with high MRMR ranking

1 ≤ i ≤ m

//Feature Encoding

for each categorical data in MS do

D = get_dummies

//Feature Scaling

for each numerical data in MS do

S = get_standardscalers

DS = joint(D,S)    //obtained dataset after feature engineering

//Dataset Split

(DS1, DS2) = split (DS, ratio = 0.7)

//split dataset into train DS1 (70%) and DS2 (30%) sets

//Model Training

Modelpop = XGB(DS1)

//Model Evaluation

for each di in DS2 do

mi = Modelpop(di)    //attain prediction of power outage duration

MXGB = MXGB Ս mi

end

//Indexing Predicted Duration

for each MXGB xi do

if xi ≤ 60 then

Return “VERY SHORT”

elseif xi ≤ 120 then

Return “SHORT”

elseif xi ≤ 240 then

Return “MEDIUM”

else

Return “LONG”

end

end

End