Description
PROBLEM DEFINITION
Computers cannot beat humans on handwriting recognition yet. Handwriting
recognition is an important research area in AI research with many applications.
See these pages for further information about the problem.
1. http://en.wikipedia.org/wiki/Handwriting_recognition
2. http://archive.ics.uci.edu/ml/datasets/PenBased+Recognition+of+Handwritten+Digits (original link to the dataset)
3. http://yann.lecun.com/exdb/mnist/ (this is a larger dataset. Check the
performance of various algorithms.)
DATA
You are given a training and test dataset.
1. Pendigits.tra.csv
2. Pendigits.tes.csv
Each row contains 16 features (a 4×4 bitmap of a handwritten digit) and its
correct label on the last column. Every feature is a numeric gray-scale pixel
value. 255 is white and 0 is black. Here is an example of a 8×8 scanned
handwritten digit:
You do not need to do any preprocessing to use this dataset.
TASK
1. Import the data to R (no cleanup necessary).
2. Use k nearest neighbor to predict the digit labels on the test dataset (you
may use a library).
3. Optimize the value of k to get the best prediction performance.
OUTPUT
A zip file containing the following
• A report with:
o Model
o Results
• All the source code
