How to improve the accuracy of prediction in scikit-learn

Question

I want to predict a parameter based on 3 features and 1 target. Here is my input file (data.csv):

feature.1   feature.2   feature.3   target
    1           1          1        0.0625
    0.5         0.5        0.5      0.125
    0.25        0.25       0.25     0.25
    0.125       0.125      0.125    0.5
    0.0625      0.0625     0.0625   1

Here is my code:

import pandas as pd
from sklearn.model_selection import train_test_split
from collections import *
from sklearn.linear_model import LinearRegression

features = pd.read_csv('data.csv')

features.head()
features_name = ['feature.1' , 'feature.2' , 'feature.3']
target_name = ['target']

X = features[features_name]
y = features[target_name]

# Split the data into training and testing sets
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.1, random_state = 42)

linear_regression_model = LinearRegression()
linear_regression_model.fit(X_train,y_train)

#Here is where I want to predict the target value for these inputs for 3 features
new_data  = OrderedDict([('feature.1',0.375) ,('feature.2',0.375),('feature.3',0.375) ])

new_data = pd.Series(new_data).values.reshape(1,-1)
ss = linear_regression_model.predict(new_data)
print (ss)

According to the trend, I expect to get a value around 0.1875 if I give 0.375 as input for all of the features. However the code predicts this:

[[0.44203368]]

Which is not correct. I do not know where the problem is. Does anybody know how I can fix it?

Thanks

Answer 1

Your data is not linear. I have plotted just one dimension since the features are identical:

Approximating a Non-Linear function by a LinearRegression model creates bad results, like you experienced. You could try to model a better fitting function and fit its parameters with scipy: https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.curve_fit.html