Author:

• Shoval Tayro

About the project:

My research question is- given a new post- whether it is possible to predict that the writer will commit suicide after the post or not.

In this project I used 5 machine learning techniques-

• Logistic Regreession
• Adaboost
• Random Forest
• KNN
• SVM

About the data set:

The dataset is a collection of posts from "SuicideWatch" and "depression" subreddits of the Reddit platform.

• 'text' - the text of the post.
• 'class' - The labal (if the writer commit suicide or not).
• The dataset contain 49% posts that end with 'suicide' and 51% that not end with suicide- 'non-suicide'.

Problems & Solutions:

• The biggest problem I encountered during the project is how to clear the text of unwanted characters/ words that can affect on the results and can impair the learning process-
Stop words like 'I', 'am', 'that', 'this' and etc.. are available in abundance in any human language. By removing these words, I remove the low-level information from our text in order to give more focus to the important information.
I used 'nltk' library to remove the stop words from the text, also I removed the punctuation and the leading/ trailing whitespace of each line in order to get clean text as possible. In addition I did stemming on the text -'PorterStemmer' from 'nltk' because stemming makes the training data more dense. It reduces the size of the dictionary two or three-fold and the learning process will work better.
I did all these steps on the text to try to create an optimal learning dataset.

• How to extract the features-
The techniques I used couldn't perform learning on text, only on numbers so I had to find a way to convert the text into numbers. I chose to use 'unicode' conversion after trying to convert to integer/double that ended without success. After the converting to Unicode I was able to learn about the text and get good results.

RESULSTS :

Logistic Regression:

---

Adaboost:

---

RANDOM FOREST:

---

KNN:

With CountVectorizer and with TD-IDF:

---

SVM :

With CountVectorizer and with TD-IDF:

---

Conclusions :

From this results it can be deduced that it is possible to predict whether the writer of the post committed suicide or not.
The average of the techniques I used is 83.5%.
The techniques that gived the highest results was Logistic Regression - 89.6%.
Logistic regression is a classification algorithm used to find the probability of event success and event failure. It is used when the dependent variable is binary(0/1, True/False, Yes/No in my case Suicide/Non-suicide) in nature. It supports categorizing data into discrete classes by studying the relationship from a given set of labelled data.
Also adaboost give me a good results - 88.2%. I used sklearn's default rules for adaboost.
KNN give the lowest results - 73.7%.
In my opnion KNN give the lowest result beacuse KNN works well with smaller dataset because it is a lazy learner. It needs to store all the data and then makes decision only at run time. So if dataset is large like I used (10,000 posts from Reddit), there will be a lot of processing which may adversely impact the performance of the algorithm.
I tried to get a better result with KNN by trying to play with the test/train size but in the end, the best results for KNN were when the test size was 0.25.

Source:

https://www.kaggle.com/nikhileswarkomati/suicide-watch