LU102762A1 - Machine learning-based and adaptive optimization method of ap selection - Google Patents
Machine learning-based and adaptive optimization method of ap selection Download PDFInfo
- Publication number
- LU102762A1 LU102762A1 LU102762A LU102762A LU102762A1 LU 102762 A1 LU102762 A1 LU 102762A1 LU 102762 A LU102762 A LU 102762A LU 102762 A LU102762 A LU 102762A LU 102762 A1 LU102762 A1 LU 102762A1
- Authority
- LU
- Luxembourg
- Prior art keywords
- decision tree
- tree
- node
- data set
- class
- Prior art date
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W48/00—Access restriction; Network selection; Access point selection
- H04W48/20—Selecting an access point
Landscapes
- Engineering & Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The present invention discloses a machine learning-based and adaptive optimization method of AP selection. The method is applied to a process of setting up WiFi connections between mobile devices and APs and a process of adaptive network handovers in the Internet of Vehicles. The method includes: collecting connected device data in the current environment, creating a training data set and a feature set, and determining a threshold; determining whether a decision tree is a single-node tree based on the data set and the ID3 algorithm; if the decision tree is not a single-node tree, obtaining subsets by division to construct subnodes and generate the tree; and perform recursive invoking until a complete decision tree is generated, classifying APs into a fast set and a slow set, and selecting the fastest AP from the fast set to set up a connection. In the present invention, access points (APs) are selected based on a machine learning model to shorten the connection time and reduce the time cost for WiFi connection setting.
Description
1 AO 21.04.1093 LU LU102762 MACHINE LEARNING-BASED AND ADAPTIVE OPTIMIZATION METHOD OF
TECHNICAL FIELD The present invention relates to the fields of communication technologies and Internet of Vehicles adaptive switching, and in particular to a machine learning-based and adaptive optimization method of AP selection.
BACKGROUND In recent years, wireless data traffic has shown an exponential increase trend due to the explosive growth of smart devices. Among wireless networks, 802.11 wireless local area networks (WiFi) have become a major part of current wireless services. Over the past decade, more than 1 billion WiFi access points (APs) have been deployed to provide wireless connectivity. Users’ smart devices that support 3G/4G cellular networks can also access ubiquitous WiFi hotspots. However, the network performance and user experience of the WiFi networks are unsatisfactory. According to the measurement report on more than 5 million WiFi network users in urban areas, up to 45% of the mobile devices cannot set up connections to WiFi APs, and the setup time of 15% (5%) of the successful WiFi connections exceeds 5 seconds (10 seconds). Most of the previous research on WiFi | networks focused on common metrics about user experience (e.g., the bandwidth and latency experienced in the WiFi networks), with little attention paid to the performance of the WiFi connection setup process. The data collected from Android smartphones in controlled environments shows that the loss of most connection setup time cost is resulted from a loss of DHCP packets. In fact, the performance of the outdoor WiFi connection setup process is still unknown, and there is a lack of more thorough investigation. Much research has concentrated on WiFi performance measurement to estimate available throughputs of some AP-client links and explore the latency of APs. Motivated by the existing research works, it is urgent to consider metrics about the connection setup time cost, because the high connection failure rate has affected the user experience. Moreover, network connection handover issues also exist in the field of Internet of
2 AO 21.04.1093 LU LU102762 Vehicles (IoV). In the prior art, when a mobile device moves within a network, the mobile device will be connected to various APs. Consequently, significant quality of service fluctuations and possible long connection interruptions appear during the handover process, and the signal power is insufficient to support high data rate: These problems usually occur in various life scenarios such as elevators and stairs, particularly in the IoV field. When the user arrives at a network blind spot, the connection is interrupted. Apparently, the data flow on the mobile device will be particularly affected by the temporary connection loss, and this situation is considered a major problem by the users.
Currently, there is little research on the connection setup and adaptive handover process. Most of the current research on the connection setup process focuses on WiFi handover mechanisms, that is, aims to reduce the handover latency. Various solutions have been proposed in the prior art to mitigate these problems to receive information about the upcoming connection loss, such as loss predictions from the device or appropriate intervention to a player. The key elements of all expected policies are the long-term predictions on appropriate channel conditions, and the time scale thereof is much larger than that of small-scale fading. For prediction, most existing methods are based on particular channel models or broad and detailed channel maps. Apparently, these two approaches may not be sufficient to meet data flow requirements.
SUMMARY The main objective of the present invention is to provide a machine learning-based and adaptive optimization method of AP selection to solve the problems in the prior art. The specific technical solutions are as follows: Disclosed is a machine learning-based and adaptive optimization method of AP selection, where the method includes the following steps: step 1: collecting data of the connected devices in the current environment, creating a training data set and a feature set, and determining a threshold; step 2: determining whether a decision tree is a single-node tree based on the data set and the ID3 algorithm; where assuming that k classes Ck (k = 1, 2, 3, ..., K) exist, if all instances in the training data set belong to the same class Ck, the decision tree is a single-node tree, and the class Ck
3 AO 21.04.1093 LU LU102762 is used as a class tag of the node to return the decision tree.
If the feature set is an empty set, the decision tree is a single-node tree, and a class Ck with the largest number of instances in the training data set is used as a class tag of the node to return the decision tree.
Otherwise, an information gain of each feature in the feature set for the training data set is calculated based on the ID3 algorithm, and a feature Ag with the largest information gain is selected.
If the information gain of Ag is less than a threshold e, the decision tree is set as a single-node tree, then a class Ck with the largest number of instances in the training data set is used as a class tag of the node to return the decision tree; step 3: if the decision tree is not a single-node tree, obtaining subsets by division to construct subnodes and generate the tree; and step 4: recursively invoking step 2 and step 3 above until a complete decision tree is generated, classifying APs into a fast set and a slow set, and selecting the fastest AP from the fast set to set up a connection.
Further, in step 1, the created feature set includes but is not limited to a time used for connection, signal strength, a mobile device model, information about whether an AP is a public AP, information about whether data is encrypted, and the number of connected devices, and available algorithms include but are not limited to decision tree algorithms such as ID3 and C4.5. Further, in step 2, steps for calculating the information gain based on the ID3 algorithm are as follows: step 2-1: calculating an empirical entropy H(D) of a data set D; LG) IG H(D)= Sl, a step 2-2: calculating an empirical conditional entropy H(D|A) of a feature A for the data set D; and 5 D; 5 DE | Di D, H(D| A)= Seh )= an 2 step 2-3: calculating an information gain g(D,A) as g(D,A) = H(D) - H(DjA). Further, in step 3, specifically, after it is determined whether the decision tree is a single-node tree, for each possible value ai of Ag, the training data set is divided into
4 AO 21.04.1093 LU LU102762 several non-empty subsets Di based on Ag = ai, a class with the largest number of instances in Di is used as a tag to construct subnodes, and the decision tree is constructed based on the node and the subnodes and then is returned. Compared with the prior art, in the application process of the present invention, the number of connection attempt failures is less than 3.6%of total attempts. The time cost is only 3 seconds for 80% of connections. In contrast, the time cost exceeds 30 seconds when a baseline algorithm is used. In other words, the time cost is reduced by 10 times for 80% of the connections. The algorithm of the present invention takes into account a high probability of connection failure events, even if the measured signal strength is the highest on the mobile devices. The model of the present invention can predict these connection failure events at higher accuracy and prevent the mobile devices from connecting to APs in the slow set.
BRIEF DESCRIPTION OF DRAWINGS FIG. 1 is a schematic flowchart of a method according to the present invention; FIG. 2 shows relative information gains of a feature set in an embodiment; and FIG. 3 shows a decision tree model generated in an embodiment.
DESCRIPTION OF EMBODIMENTS The following describes the technical solutions of the present invention in further detail with reference to the accompanying drawings of the specification. Disclosed is a machine learning-based and adaptive optimization method of AP selection, where the method includes the following steps: Step 1: Collect data of connected devices in the current environment, create a training data set and a feature set, and determine a threshold. In step 1, the created feature set includes but is not limited to a time used for connection, signal strength, a mobile device model, information about whether an AP is a public AP, information about whether data is encrypted, and the number of connected devices, and available algorithms include but are not limited to decision tree algorithms such as ID3 and C4.5. Specifically, the input includes a training data set D, a feature set A (a time used for connection, signal strength, a mobile device model, information about whether an AP is
AO 21.04.1093 LU LU102762 a public AP, information about whether the data is encrypted, the number of connected devices, and the like), and a threshold £. The output includes a decision tree T.
To understand how each function affects the connection time cost during data set creation, the present invention shows connection time cost differences between the functions 5 through coordinate axis visualization.
As there are thousands of different mobile device models and AP models, the embodiments omit the coordinate axis visualization results of the functional mobile device models and AP models.
In an embodiment, relative information gains of some typical features are selected, as shown in FIG. 2. Step 2: Determine whether a decision-tree is a single-node tree based on the data set and the ID3 algorithm.
Assuming that k classes Ck (k = 1, 2, 3, ..., K) exist, if all instances in the training data set belong to the same class Ck, the decision tree is a single-node tree, and the class Ck is used as a class tag of the node to return the decision tree; if the feature set is an empty set, the decision tree is a single-node tree, and a class Ck with the largest number of instances in the training data set is used as a class tag of the node to return the decision tree.
Otherwise, an information gain of each feature in the feature set for the training data set is calculated based on the ID3 algorithm, and a feature Ag with the largest information gain is selected.
If the information gain of Ag is less than the threshold €, the decision tree is set as a single-node tree, then a class Ck with the largest number of instances in the training data set is used as a class tag of the node to return the decision tree.
In step 2, steps for calculating the information gain based on the ID3 algorithm are as follows: Step 2-1: Calculate an empirical entropy H(D) of a data set D.
H(D)= 1G] gg, lc] “of ip] Step 2-2: Calculate an empirical conditional entropy H(D|A) of a feature A for the data set D.
AD 2 DED, D H(D| 2-30) gos bel, x Step 2-3: Calculate an information gain g(D,A)as g(D,A) = H(D) - H(DJA).
6 AO 21.04.1093 LU LU102762 Step 3: If the decision tree is not a single-node tree, obtain subsets by division to construct subnodes and generate the tree. In step 3, specifically, after it is determined whether the decision tree is a single-node tree, for each possible value ai of Ag, the training data set is divided into several non- empty subsets Di based on Ag = ai, a class with the largest number of instances in Di is used as a tag to construct subnodes, and the decision tree is constructed based on the node and the subnodes and then is returned.
Step 4: Recursively invoke step 2 and step 3 above until a complete decision tree is generated, classify APs into a fast set and a slow set, and select the fastest AP from the fast set to set up a connection. For the i" subnode, Di is used as the training set and A—{Ag} is used as the feature set to perform recursive invoking. A subtree Ti is obtained and then is returned. The decision tree model generated in the embodiment is shown in FIG. 3. Compared with the prior art, in the application process of the present invention, the number of connection attempt failures is less than 3.6%of total attempts. The time cost is only 3 seconds for 80% of connections. In contrast, the time cost exceeds 30 seconds when a baseline algorithm is used. In other words, the time cost is reduced by 10 times for 80% of the connections. The algorithm of the present invention takes into account a high probability of connection failure events, even if the measured signal strength is the highest on the mobile devices. The model of the present invention can predict these connection failure events at higher accuracy and prevent the mobile devices from connecting to APs in the slow set. The foregoing description is merely example embodiments of the present invention, and the protection scope of the present invention is not limited to the embodiments described above. However, any equivalent modification or change made by a person of ordinary skill in the art based on the disclosure of the present invention shall fall within the protection scope described in the claims. |
Claims (4)
1. A machine learning-based and adaptive optimization method of AP selection, wherein the method comprises the following steps: step 1: collecting connected device data in the current environment, creating a training data set and a feature set, and determining a threshold; step 2: determining whether a decision tree is a single-node tree based on the data set and the ID3 algorithm; wherein assuming that k classes Ck (k = 1, 2, 3, ..., K) exist, if all instances in the training data set belong to the same class Ck, the decision tree is a single-node tree, and the class Ck is used as a class tag of the node to return the decision tree; if the feature set is an empty set, the decision tree is a single-node tree, and a class Ck with the largest number of instances in the training data set is used as a class tag of the node to return the decision tree; or otherwise, an information gain of each feature in the feature set for the training data set is calculated based on the ID3 algorithm, and a feature Ag with the largest information gain is selected, and if the information gain of Ag is less than a threshold e, the decision tree is set as a single-node tree, and a class Ck with the largest number of instances in the training data set is used as a class tag of the node to return the decision tree; step 3: if the decision tree is not a single-node tree, obtaining subsets by division to construct subnodes and generate the tree; and step 4: recursively invoking step 2 and step 3 above until a complete decision tree is generated, classifying APs into a fast set and a slow set, and selecting the fastest AP from the fast set to set up a connection.
2. The machine learning-based and adaptive optimization method of AP selection according to claim 1, wherein in step 1, the created feature set comprises but is not limited to a time used for connection, signal strength, a mobile device model, information about whether an AP is a public AP, information about whether data is encrypted, and the number of connected devices, and available algorithms comprise but
8 AO 21.04.1093 LU LU102762 are not limited to decision tree algorithms such as ID3 and C4.5.
3. The machine learning-based and adaptive optimization method of AP selection according to claim 1, wherein in step 2, steps for calculating the information gain based on the ID3 algorithm are as follows: step 2-1: calculating an empirical entropy H(D) of a data set D; SG, 1G H(D)=- [Gl LE (O)= pl To step 2-2: calculating an empirical conditional entropy H(DJA) of a feature A for the data set D; and > |Di D&E Pu, Pal H(D|4)=$ Par (0) = 3: 1205 Puli Da fa 12) fut |p| kat To} ID] step 2-3: calculating an information gain g(D,A) as g(D,A) = H(D) — H(D|A).
4. The machine learning-based and adaptive optimization method of AP selection according to claim 1, wherein in step 3, specifically, after it is determined whether the decision tree is a single-node tree, for each possible value ai of Ag, the training data set is divided into several non-empty subsets Di based on Ag = ai, a class with the largest number of instances in Di is used as a tag to construct subnodes, and the decision tree is constructed based on the node and the subnodes and then is returned.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| LU102762A LU102762B1 (en) | 2021-04-08 | 2021-04-08 | Machine learning-based and adaptive optimization method of ap selection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| LU102762A LU102762B1 (en) | 2021-04-08 | 2021-04-08 | Machine learning-based and adaptive optimization method of ap selection |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| LU102762A1 true LU102762A1 (en) | 2021-10-08 |
| LU102762B1 LU102762B1 (en) | 2022-03-11 |
Family
ID=75478138
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| LU102762A LU102762B1 (en) | 2021-04-08 | 2021-04-08 | Machine learning-based and adaptive optimization method of ap selection |
Country Status (1)
| Country | Link |
|---|---|
| LU (1) | LU102762B1 (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120170471A1 (en) * | 2010-12-31 | 2012-07-05 | Openpeak Inc. | Automated access point selection to provide communication network presence to a communication device |
| US20150139074A1 (en) * | 2013-11-15 | 2015-05-21 | Ryan H. Bane | Adaptive Generation of Network Scores From Crowdsourced Data |
| US20160278007A1 (en) * | 2015-03-20 | 2016-09-22 | Qualcomm Incorporated | Selection of an access point in a wireless communications network |
| US20170126705A1 (en) * | 2015-10-29 | 2017-05-04 | Mojtaba Mojy Mirashrafi | Wireless hotspot attack detection |
-
2021
- 2021-04-08 LU LU102762A patent/LU102762B1/en active IP Right Grant
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120170471A1 (en) * | 2010-12-31 | 2012-07-05 | Openpeak Inc. | Automated access point selection to provide communication network presence to a communication device |
| US20150139074A1 (en) * | 2013-11-15 | 2015-05-21 | Ryan H. Bane | Adaptive Generation of Network Scores From Crowdsourced Data |
| US20160278007A1 (en) * | 2015-03-20 | 2016-09-22 | Qualcomm Incorporated | Selection of an access point in a wireless communications network |
| US20170126705A1 (en) * | 2015-10-29 | 2017-05-04 | Mojtaba Mojy Mirashrafi | Wireless hotspot attack detection |
Also Published As
| Publication number | Publication date |
|---|---|
| LU102762B1 (en) | 2022-03-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10820244B2 (en) | Cell broadcast for smart traffic steering across radio technologies with improved radio efficiency | |
| US11265205B2 (en) | System and method for anomaly detection with root cause identification | |
| US7864727B2 (en) | Congestion control method for wireless communication system and base station control apparatus | |
| US7653392B2 (en) | Methods and systems for heterogeneous wireless network discovery and selection | |
| US9591536B2 (en) | Cell user occupancy indicator to enhance intelligent traffic steering | |
| He et al. | A simple and robust vertical handoff algorithm for heterogeneous wireless mobile networks | |
| US8520639B2 (en) | Apparatus and method for performing handover in advanced mobile communication system | |
| US11109283B1 (en) | Handover success rate prediction and management using machine learning for 5G networks | |
| US20160286410A1 (en) | Techniques for managing handovers, ncls, and connections in a radio network | |
| US10560864B2 (en) | Event-driven network demand finder of a radio access network | |
| US9900845B2 (en) | Battery saving with radio control based on cellular condition data | |
| CN103959850A (en) | Measurements for the interworking of wireless wide area and wireless local area networks | |
| KR20120062814A (en) | Methods for managing co-located macro and femto base station deployments and methods for initiating mobile station handoff | |
| Cheelu et al. | A study of vertical handoff decision strategies in heterogeneous wireless networks | |
| CN116671068A (en) | Policy determination method and device | |
| Alraih et al. | Effectiveness of handover control parameters on handover performance in 5G and beyond mobile networks | |
| Satapathy et al. | An efficient multicriteria‐based vertical handover decision‐making algorithm for heterogeneous networks | |
| CN110009061B (en) | AP self-adaptive optimization selection method based on machine learning | |
| EP1677556A1 (en) | Method and System for Operating a Mobile Communications Network | |
| US20240012675A1 (en) | Dynamic discovery mechanism in 5g systems and methods | |
| LU102762A1 (en) | Machine learning-based and adaptive optimization method of ap selection | |
| Becvar | Efficiency of Handover Prediction Based on Handover History. | |
| AT&T | vodi.dvi | |
| Venkatesh et al. | Optimizing handover in LTE using SON system by handling mobility robustness | |
| Wu et al. | A crossing-tier location update/paging scheme in hierarchical cellular networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| FG | Patent granted |
Effective date: 20220311 |