Machine Learning Assignment

1) Regression; Consider the Bike Share dataset from the UCI machine learning repository. The dataset contains three files, viz. day.csv, hour.csv, and Readme.txt. Both the datasets (data.csv and hour.csv) contain a combination of integer-valued (e.g., season, weekday or not) and real-valued features (e.g., temperature, windspeed). Details of the dataset are described in the README (Readme.txt) in the Data Characteristics section. Just like the previous assignment, spend some time understanding the structure of the dataset, how the instances are organized, how the features are organized, what the various features mean (info in README), what features are useful for the task at hand, and so on. Do not attempt to run any machine learning algorithm before understanding the structure of the dataset.

Note, in particular, the last three fields in the data, viz. casual (denotes casual riders), registered (registered riders), and cnt (total ridership count).

1. Using only hour.csv, implement regression algorithms (both linear and k-nearest neighbors) to predict the hourly values for:

a. the number of casual riders

b. the number of registered riders

c. total ridership count

2. Using only day.csv, implement regression algorithms (both linear and k-nearest neighbors) to predict the daily values for:

a. the number of casual riders

b. the number of registered riders

c. total ridership count

Therefore, for each dataset, you are reporting 3 models for linear regression and 3 models for KNN regression.

Note: Remember that using one of the target values (as a feature) in predicting the outcome of any of the counts - casual, registered, or total would defeat the purpose of the learning algorithm. It will make the problem too easy. That is, the number of casual riders cannot be used as a feature to predict the number of registered riders or total ridership. Similarly, the number of registered riders cannot be used to predict the other two values, and so on. Only features like season, temperature, real-feel, etc. have to be used by the learning algorithm.

For instance, suppose you are using hourly.csv dataset and you want to predict the number of casual riders. You have to remove the columns related to the number of registered riders and total ridership first and then start training/testing your model. Similarly, when you are making the prediction model for total ridership, you have to remove the columns related to casual rides and registered riders first and then start training/testing your model.

As before, you will need to separate the data into training set and test set (decide on the proportion of splits yourself). Evaluate the performance of your regression using suitable measures. Report on the performance results and which model(s) worked best (and why in your opinion).

2) Clustering; Consider the Seeds data set from the UCI machine learning repository. The dataset comprises of features from three different types of wheat kernels. There are seven features (area, perimeter, compactness, length, width, asymmetry coefficient, and length of kernel groove) that describe each data point. (Note that the dataset has an eighth column (class information with labels 1, 2, and 3), which we will use as ground truth to verify our clustering results.)

Using the k-means algorithm cluster this dataset into three clusters based on the seven features at your disposal. Demonstrate the effectiveness of your implementation by comparing the results against the ground truth. Follow the steps in the k-means demo video from the lectures.

Also, note that the default label values in scikit learn start from 0, whereas the dataset here starts labels with 1. While evaluating your implementation's effectiveness, ensure to account for this discrepancy.

As a performance measure, compare the clusters identified by k-means w.r.t. the ground truth data and make observations.

Attachment:- Assignment Files.rar