Safe Route - Machine Learning

Overview

• Clustering
• Habit Tracker
• Known Issues
• Team

Clustering

Model Description

Clustering model is produced using scikit-learn DBScan class. Centroid comes from the average of each cluster with the range is the maximum distance of centroid to one of cluster members.

Model generation

• Model : Use create_model() method
• Statistic : Use generate_area_statistic(data) with data being the DataFrame of csv file

Centroid

JSON File Format

{
    "centroids": [
        {
            "id": ... (int),
            "latitude": ... (float),
            "longitude": ... (float),
            "range": ... (float),
            "crime_info": {
                <crime_type>(string): ... (int),
                ...
            }
        },
        ...
    ]
}

Statistic

JSON File Format

{
    "statistic": [
        "subdistrict": ... (string),
        "total_crime": ... (int),
        "crime_info": {
            <crime_type>(string): ... (int),
            ...
        }
        "coordinates": ... (list of* float)
    ]
}

Statistic JSON File Note:

• coordinates list order is latitude, longitude
• list may contain another list
• please take a look at area_statistic.json before use

Habit Tracker

Model Description

Habit tracker model is created for each user that use the application, it utilize LSTM mechanics to predict multivariate multilabel time series data. Model will be fed with multiple inputs which is date, time, and location. The model will give location (latitude and longitude) as output.

Model Generation

User's model will be generated when the endpoint /create called with username as URL parameters*.

* further endpoint detail will be given below

Model

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 lstm (LSTM)                 (None, 15, 64)            18176     
                                                                 
 lstm_1 (LSTM)               (None, 15, 32)            12416     
                                                                 
 lstm_2 (LSTM)               (None, 16)                3136      
                                                                 
 flatten (Flatten)           (None, 16)                0         
                                                                 
 dense (Dense)               (None, 16)                272       
                                                                 
 dense_1 (Dense)             (None, 8)                 136       
                                                                 
 dense_2 (Dense)             (None, 2)                 18        
                                                                 
=================================================================
Total params: 34,154
Trainable params: 34,154
Non-trainable params: 0
_________________________________________________________________

• Model input shape=(15, 6)
• Model input data columns:

column	type	range	description
year	integer	*	year timestamp
month	integer	1 - 12	month timestamp
day_of_week	integer	1 - 7	day of week
time	integer	0 - 1440	cumulative minute of day
latitude	float	-90.0 - 90.0	latitude
longitude	float	-180.0 - 180.0	longitude

• Model output:

column	type	range	description
latitude	float	-90.0 - 90.0	latitude
longitude	float	-180.0 - 180.0	longitude

Flask

url: https://model-ck44nnq7hq-as.a.run.app

endpoint	url_param	body	method
/create	username(string)	-	GET
/train	username(string)	json-object-1*	POST
/forecast	username(string)	json-object-2*	POST

* refer to the format below

• json-object-1

{
	"username": ...,
	"data": [
	        {
	            "datetime": ...,
	            "latitude": ...,
	            "longitude": ...
	        },
	        ...
	        {
	            "datetime": ...,
	            "latitude": ...,
	            "longitude": ...
	        }
	    ]
}

• json-object-2

{
	"username": ...,
    "email": ...,
    "latitude": ...,
	"longitude": ...,
	"data": [
	        {
	            "datetime": ...,
	            "latitude": ...,
	            "longitude": ...
	        },
	        ...
	        {
	            "datetime": ...,
	            "latitude": ...,
	            "longitude": ...
	        }
	    ]
}

• json key-value description

key	value	range
username	string	-
email	string	-
latitude	float	-90.0 - 90.0
longitude	float	-180.0 - 180.0
data	json-array-of-object	-
datetime	timestamp	%yyyy/%mm/%dd %hh:%MM:%ss

Known Issues

• Clustering - Require real data from local authorities
• Habit Tracker - Model architecture has not yet been optimized
• Habit Tracker - Costly data and model pipelining
• Habit Tracker - Require a lots of data to create optimum model

Team

• M2002F0101 - Christopher Chandrasaputra
• M2006F0613 - Hanif Adam Al Abraar
• A2183F1771 - Fillah Akbar Firdausyah
• A2009J0968 - Ikhsan Cahya Mardika
• C2009F0967 - I Putu Cahya Adi Ganesha
• C2306G2617 - Muhammad Anggi Wirahmat