Based on the dataset, this algorithm predicts the possibility of low birth weight of a child given the following set of features:

• Age of the Mother in Years
• Weight in Pounds at the Last Menstrual Period
• Race (1 = White, 2 = Black, 3 = Other)
• Smoking Status During Pregnancy (1 = Yes, 0 = No)
• History of Premature Labor (0 = None, 1 = Yes)
• History of Hypertension (1 = Yes, 0 = No)
• Presence of Uterine Irritability (1 = Yes, 0 = No)

Using this feature set (x0, x1 .. x6), higher order features were generated to get greater accuracy of prediction. I used Logistic Regression along with python modules scipy and numpy for this purpose. After training on the dataset, I predicted the Low Birth Weight (0 = No, 1 = Yes) values for the same data-set and compared it against the actual values. With this the accuracy of prediction was calculated. With feature mapping and regularization the algorithm achieved an accuracy of 82.14%.

    Hosmer and Lemeshow (2000) Applied Logistic Regression: Second
    Edition.  These data are copyrighted by John Wiley & Sons Inc. and must
    be acknowledged and used accordingly.  Data were collected at Baystate
    Medical Center, Springfield, Massachusetts.

Look at the data_description file for more details.

• Hosmer and Lemeshow for providing the dataset.