- Predicting Daily Activities and Sports Project
This project provides a tuning process on the different sets of parameters in predicting „Daily and Sports Activities‟, human activities performed while wearing sensor units on the chest, arms, and legs. The dataset comprises motion sensor data of 19 different daily activities and sports each performed by 8 subjects in their own style for 5 minutes. The combination of layers with high densities, dropouts, and optimizers in two layers neural network are considered. The optimal model in this project is the model with 1406 nodes in each layer and dropout of 0.3, with 93% of accuracy.
The dataset is not available
- Predicting the Amount of Active Ingredient of Pharmaceutical Tablets Data
A regularized logistic regression model with L1 penalty function is used to predict the amount of active ingredient given the input near infrared spectrometry (NIR) measurements of pharmaceutical tablets data.
The dataset is available
- Predicting handwritten Digits from the USPS
A multilayer neural network is used to predict the digit type using 265 input features corresponding to the grid representation of the images. The file contains data recording handwritten digits from the United States Postal Service (USPS). More in details, the data consists of grayscale (16x16) grid representations of image scans of the digits “0” through “9” (10 digits). The data is divided into train and test data, containing respectively 7291 and 2007 observations.
The dataset is available