Predictive Models for Ebola using Machine Learning in Python

Output:-

array(['Guinea', 'Liberia', 'Sierra Leone', 'United Kingdom', 'Mali',
       'Nigeria', 'Senegal', 'Spain', 'United States of America', 'Italy',
       'Liberia 2', 'Guinea 2'], dtype=object)

T_Guinea=[73.0,84.0,78.5]
T_Liberia=[81.0,81.0,81.0]
T_Sierra_Leone=[78.6,83.3,80.95]
T_United_Kingdom=[40.0,60.0,50.0]
T_Mali=[77.2,90.3,83.75]
T_Nigeria=[78.8,82.4,80.6]
T_Senegal=[70.0,82.0,76.0]
T_Spain=[46.0,73.0,59.5]
T_United_States_of_America=[33.2,71.9,52.55]
T_Italy=[46.0,78.0,62.0]

T = [T_Guinea, T_Italy, T_Liberia, T_Mali, T_Nigeria, T_Senegal,
     T_Sierra_Leone, T_Spain, T_United_Kingdom, T_United_States_of_America]

print(T[0][0])

73.0

ntotal = 9947
ntr = 8000
nts = ntotal - ntr
df_tot = df.loc[0:ntotal,:]

Preprocessing

x_df = df_tot.loc[df['Indicator'] == 'Cumulative number of confirmed, probable and suspected Ebola cases', ['Country', 'Date', 'value']]

x_df_country = x_df.take([0], axis=1)
x_df_date = x_df.take([1], axis=1)

x_country = x_df_country.values
x_date = x_df_date.values
x_date = x_date.ravel()
x_days = pd.to_datetime(x_date, format= "%Y-%m-%d").dayofyear.values
x_country = x_country.ravel()

Now extract the confirmed Ebola cases:-

ls = []
for i in range(x_date.shape[0]):
    ls.append(df_tot.loc[(df['Date'] == x_date[i]) & 
                        (df['Country'] == x_country[i]) & 
                        (df['Indicator'] == 'Cumulative number of confirmed Ebola cases'),
                        ['value']].take([0], axis=1).values)

Convert cd list to NumPy array of float elements:-

cd_arr = np.array(ls)
for i in range(len(cd_arr)):
    if cd_arr[i].ravel().size == 0:
        cd_arr[i] = float(0.0)
    try:
        cd_arr[i] = cd_arr[i].ravel()[0]
    except:
        continue

Get the total Ebola cases:-

totd_arr = x_df.take([2], axis=1).values
totd_arr = totd_arr.ravel()

Compute the percentage of confirmed ebola ratio but first, fix invalid fields

invalid_fields = np.where(totd_arr == 0)[0] 
totd_arr[invalid_fields] = (1, 1) 
cd_arr[invalid_fields] = (0, 0) 
cd_ratio = cd_arr / totd_arr

y = cd_ratio

Create three arrays ready to be replaced by the lowest temperature,

highest temperature and average temperature.

x_lo_T=np.array([i for i in x_country])
x_hi_T=np.array([i for i in x_country])
x_avg_T=np.array([i for i in x_country])

nCountry = len(np.unique(x_country))

for c in range (nCountry):
    np.place(x_lo_T, x_lo_T == np.unique(x_country)[c], T[c][0])
for c in range (nCountry):
    np.place(x_hi_T, x_hi_T == np.unique(x_country)[c], T[c][1])
for c in range (nCountry):
    np.place(x_avg_T, x_avg_T == np.unique(x_country)[c], T[c][2])

X=np.row_stack([xtr_lo_T, xtr_hi_T, xtr_avg_T, x_days]).T
X=X.astype(float)
# Normalization
X=X/np.max(X,axis=1)[:,None]
X.shape

Linear Model

Linear models describe a continuous response variable as a function of one or more predictor variables. They can help you understand and predict the behavior of difficult systems or analyze experimental, financial, and natural data.

We import the linear model and use the first 80% of data for training and the rest for analysis.

from sklearn import linear_model
ntr = int(np.floor(0.8*X.shape[0]))
xtr = X[:ntr,:]
ytr = y[:ntr]
xts = X[ntr:, : ]
yts = y[ntr:]
#linear regression
regr = linear_model.LinearRegression()
#Trainnig model
regr.fit(xtr,ytr)

#Prediction
y_pred = regr.predict(xts)

Therefore, Assume countries and days are correlated,

Xtr = np.column_stack((np.ones(ntr),xtr[:,0],xtr[:,1],xtr[:,2],xtr[:,3]))
Xts = np.column_stack((np.ones(xts.shape[0]),xts[:,0],xts[:,1],xts[:,2],xts[:,3])) 
w1 = ytr[None,:,None]/Xtr[None,:,:] 
w1 = np.mean(W1,axis=1) 
y_pred1 = w1*Xts

import matplotlib
import matplotlib.pyplot as plt

plt.plot(y_pred,yts,'.')
plt.plot([0.25,1.5],[0.25,1.5],'r-',linewidth=1)

Output Graph:-

Conclusion:

Further ,”The countries with higher temperature (room temperature or above) is likely to increase the speed at which the virus dies in the atmosphere.”

Refrences: https://www.who.int