import pandas as pd
df = pd.read_csv("fake_news_dataset.csv") #reading the dataset 
df = df[['id', 'state', 'date_published', 'source',                       
       'category', 'sentiment_score', 'word_count', 'has_images',
       'has_videos', 'readability_score', 'num_shares', 'num_comments',
       'political_bias', 'fact_check_rating', 'is_satirical', 'trust_score',
       'source_reputation', 'clickbait_score', 'plagiarism_score', 'label']]   #we dont want author, title, character count or text because the
                                                                               #information is irrelevant, so we remove it

df.dropna #we drop all NA values

df['True'] = df['label'].replace({'Fake':0, 'Real': 1}) #we make a new column called "True" and changing the string "Fake" to 0 and "Real" to 1.

df = df[['id', 'state', 'date_published', 'source',
       'category', 'sentiment_score', 'word_count', 'has_images',
       'has_videos', 'readability_score', 'num_shares', 'num_comments',
       'political_bias', 'fact_check_rating', 'is_satirical', 'trust_score',
       'source_reputation', 'clickbait_score', 'plagiarism_score', 'True']]   #we do the same thing, we just remove  'label'
  
df.head(4)  #print out the first four samples

C:\Users\leoxi\AppData\Local\Temp\ipykernel_6828\3749520299.py:12: FutureWarning: Downcasting behavior in `replace` is deprecated and will be removed in a future version. To retain the old behavior, explicitly call `result.infer_objects(copy=False)`. To opt-in to the future behavior, set `pd.set_option('future.no_silent_downcasting', True)`
  df['True'] = df['label'].replace({'Fake':0, 'Real': 1}) #we make a new column called "True" and changing the string "Fake" to 0 and "Real" to 1.

	id	state	date_published	source	category	sentiment_score	word_count	has_images	has_videos	readability_score	num_shares	num_comments	political_bias	fact_check_rating	is_satirical	trust_score	source_reputation	clickbait_score	plagiarism_score	True
0	1	Tennessee	30-11-2021	The Onion	Entertainment	-0.22	1302	0	0	66.18	47305	450	Center	FALSE	1	76	6	0.84	53.35	0
1	2	Wisconsin	02-09-2021	The Guardian	Technology	0.92	322	1	0	41.10	39804	530	Left	Mixed	1	1	5	0.85	28.28	0
2	3	Missouri	13-04-2021	New York Times	Sports	0.25	228	0	1	30.04	45860	763	Center	Mixed	0	57	1	0.72	0.38	0
3	4	North Carolina	08-03-2020	CNN	Sports	0.94	155	1	0	75.16	34222	945	Center	TRUE	1	18	10	0.92	32.20	0

Questions we want to answer using this data

1. Which state might be populated with the most fake news
2. What is the difference between real and fake news and how can the model can detect it?
3. Are there common topics that are fake or real among the data?
4. What is the relationship of increase in fake news when a political event is occuring?

Variables we will be using

If you look at the dataset above, we will be using all of these variables. Here are them in a list:

'id' : Number label of the news
'state' : where the news was published
'date_published' : date when the article was published
'source' : where the article was published
'category' : type of article
'sentiment_score' : emotion score
'word_count' : word count of article
'has_images' : if the article has images
'has_videos' : if the article has vidoes
'readability_score' : how easy the text is to read
'num_shares' : how many times the article was shared
'num_comments' : how many comments were left
'political_bias' : if the article was more republican or democratic
'fact_check_rating' : if the article was fact checked
'is_satirical' : if the article is satire
'trust_score' : trustworthiness of the article
'source_reputation' : reputability of the source
'clickbait_score' : how many people clicked on the article
'plagiarism_score' : percentage of work matches sources in plagiarism database
'True' : if the article is fake or real

Methods we will use to answer our questions

We will use different modeling methods to find out which one gives us a more precise answer, such as using Logistic Regression, Decision Trees, SVMs, kNN, etc. We also want to use train_test_split in order to train our model on "new" data, to ensure there isn't any bias, and it is handling the information well.

We will also use model data to capure it's accuracy, precision recall, w coefficients and w intercepts to determine accuracy and desired fitting.