Unit 7 Project by Kendal Chapin

Machine Learning

kendalchapin — 2025-05-14 16:41:59 UTC

Machine learning is a study that uses computers systems, artificial intelligence, and human bias to perform tasks without specific instructions from the user.

Ex: Spotify uses machine learning by taking data from the user's music history and taste, as well as the history and music style of the user's friends, in order to recommend songs that the user will like, without the user having to search them up.

Content-Based Filtering vs Collaborative Filtering

kendalchapin — 2025-05-14 16:47:51 UTC

Content-based filtering: takes the user's data and uses it to recommend songs, products, etc. to the user

Collaborative filtering: uses data from the user's friends to recommend music or shopping items to the user

Content-Based Filtering Breakdown

kendalchapin — 2025-05-14 16:57:48 UTC

Content-based filtering is made up of a series of steps:

Input(s): the attributes (individual measurable properties) and your personal ratings of a song or other website are submitted into the machine as data for study
Training data: used by the machine to create the model
Testing data: used by the machine to see if the model works
Output: the machine then takes these two types of data and uses them to make a prediction of what your rating would be

Song Rating

kendalchapin — 2025-05-14 17:07:25 UTC

The purpose of listening to the 23 songs and assigning ratings was to observe and analyze the machine learning process. By rating our songs, we were able to see how the machine responds to that data. When the machine makes its prediction of what our rating for a particular song would be, we are then able to analyze the computer's accuracy. This analysis allows us to think of ways that the machine could improve its model.

Reasons for Choosing My Top 3 Attributes:

kendalchapin — 2025-05-14 17:19:19 UTC

Tempo: According to the model, the higher the tempo, the higher my rating of the song is

Popularity: There is a strong correlation between the popularity of a song and my individual rating of the song

Acousticness: According to the model, there is a strong correlation between acousticness and a higher rating

Choosing Your Level of Complexity

kendalchapin — 2025-05-14 17:32:38 UTC

We choose our level of complexity for our training and testing models based on how well the data fits the shape of the graph.

Ex: In lesson 7.4, we tried to predict the wingspan of a student by using their height. The best fitted line for the data was a cubic level of complexity. This means that the cubic line had the least amount of error, unlike a quadratic or quartic line, which didn't fit the points as well.

One Attribute Model

kendalchapin — 2025-05-14 17:41:04 UTC

The one attribute model predicted that I would rate the song a 4.67. This prediction was very accurate because my rating of the song was a 4.

Two Attribute Model

kendalchapin — 2025-05-14 18:53:54 UTC

The two attribute model predicted that I would rate the song a 6.67. This prediction is less accurate than the one attribute model because it is higher than my actual rating, which was a 4.

Class Rating Matrix

kendalchapin — 2025-05-15 00:33:13 UTC

The students in our class who had the most similar ratings to mine were Lilly Barrett and Rymill Bucago.

The probability that Lilly likes the song if I like the song is 100%.

The probability that Rymill likes the song if I like the song is also 100%.

Recommender System Results

kendalchapin — 2025-05-15 00:50:26 UTC

Compared to my actual ratings, the results are accurate by about 50%. In other words, half of the predictions were fairly accurate, while the other half were way off.

One prediction the machine made that I was surprised about wass "We're Good" by Dua Lipa. Based on my attributes, I assumed that it would predict my rating to be high. However, it predicted that I would rate the song way lower than I did. I think the primary reason why this prediction was inaccurate is because almost all of the students gave the song a low rating, so the machine assumed that I would give it a low rating as well. I also think this occurred because my other attributes, like tempo, signaled to the machine that I prefer songs with an upbeat feel. Since this song has a smooth, calm tempo, the machine predicted I would not like it as much.

From these predictions, I have learned that collaborative filtering is not very reliable because all of the participants in an experiment are different, which means the machine can't assume everyone has the same taste or bias.

Reflection

kendalchapin — 2025-05-15 01:09:12 UTC

The most important information for the general public to know about machine learning is that it can be very helpful, but it is not reliable. Even though the machine made some very accurate predictions based on the training data, it was also very inaccurate in other areas due to a strong focus on collaborative filtering. Bias will always exist within the system because every participant in a study is unique and unpredictable.

A few different types of bias that are recognizable in machine learning include social desirability and selection bias. Social bias can occur when a listener doesn't like a song, but pretends to like it to make themselves look good. Selection bias may have occurred because the participants selected to be in this study were different from individuals who didn't take part in the study.